Welcome! In this session, we’ll explore how you can integrate and customize your own AI models within the Uptiq AI Workbench to build powerful, agentic workflows.

Jump into the exciting world of AI with UPTIQ AI Platform and start building powerful agents today! To sign up for the UPTIQ AI Workbench, follow these steps:

1. Access the Platform: Visit https://console.uptiq.ai.

1. Choose Your Sign-Up Method: Enter your preferred email address to register for the Developer Edition, or select the Sign Up with Google option for a quicker process.

2. Email Verification (if applicable): If you signed up using an email address, you will receive an invitation link in your inbox.

1. Verify Your Account: Open your email and click on the ‘Verify Email’ button to confirm and activate your account.

Once verified, you can log in and start leveraging the UPTIQ AI Workbench to build and deploy AI agents efficiently.

UPTIQ AI Workbench is an advanced platform powered by cutting-edge generative AI technology that enables financial institutions to develop and implement bespoke AI agents. These tailored agents possess the capability to understand and interpret natural language, automate a diverse array of tasks, and extract and deliver valuable insights, thereby addressing and resolving a wide spectrum of use cases effectively.

Designed to be versatile and adaptable, the UPTIQ AI Workbench is not limited to specific domains; however, it excels in offering meticulously crafted solutions particularly well-suited for the intricate needs of the enterprise financial services sector.

By leveraging the Workbench, organizations can optimize operational efficiencies, enhance decision-making processes, and ultimately generate a competitive edge within the market. The platform's ability to seamlessly integrate into existing systems further ensures a smooth transition and maximized productivity.

Consequently, businesses utilizing the UPTIQ AI Workbench are well-equipped to navigate the complexities of the financial landscape with innovative, state-of-the-art technological tools.

Common Features of UPTIQ AI Workbench:

1. Intuitive Agent Builder: The UPTIQ AI Workbench's Agent Builder allows developers to create intelligent workflows that automate repetitive back-office tasks by combining research, content generation, and software integrations. This creates a unified data foundation and streamlines operations.

2. Seamless Integration with Existing Systems: UPTIQ's Workbench provides a seamless integration experience, allowing you to effortlessly connect with any desired third-party system within your agent environment. By utilizing a declarative, low-code approach, Workbench simplifies the integration process with SaaS applications, enabling efficient data collection and streamlined processes.

Key Features of UPTIQ AI Workbench:

1. Verticalized AI Workflows: The platform offers highly specialized AI workflows that efficiently solve complex problems across wealth management, banking, and fintech sectors. This vertical integration ensures that the solutions are tailored to the unique challenges of each sector.

2. High Accuracy with No Hallucinations: UPTIQ's AI is built to avoid common issues found in large language models, such as generating incorrect or misleading outputs. By combining Retrieval-Augmented Generation (RAG) and large language models (LLMs), the platform ensures trustworthy AI outputs, which is crucial for maintaining compliance and trust in financial services.

3. Financial Data Gateway Integration: The platform features a proprietary Financial Data Gateway that connects with over 100 software platforms housing enterprise data for financial institutions' clients. This integration delivers secure, compliant, and traceable outputs, ensuring that institutions can confidently automate processes and deliver smarter, data-driven solutions.

What is PII Masking?

PII (Personally Identifiable Information) Masking is a critical feature in UPTIQ AI Workbench that ensures sensitive user data is protected during interactions with AI agents. It identifies and masks PII data in user queries before the data is passed to any LLM (Large Language Model) or other parts of the workflow. This feature allows developers to comply with data privacy regulations and build secure AI agents.

How does it work?

1. PII Recognition:

   • The PII Masking feature uses pre-built patterns to recognize common types of PII, such as:

     • Address: Detects physical addresses in the text.

Key Features

• Pre-Built Patterns: Quickly enable predefined PII recognition for commonly used data types.

• Customizable: Define custom patterns for specific business or domain needs.

• Low-Code Implementation: Activate PII Masking with minimal effort via a user-friendly interface.

• Compliance: Helps adhere to data protection regulations like GDPR, CCPA, and HIPAA.

Where to use PII Masking?

How to use PII Masking?

1. Enable Pre-Built Patterns:

   • Toggle on the required PII patterns (e.g., Address, Email, SSN) from the PII Masking interface.

2. Create Custom Patterns:

   • Use the "Create Pattern" option to define new rules for identifying PII.

Key takeaways for developers

✅ Data Security: Prevents sensitive data from being mishandled or exposed.

✅ Privacy Compliance: Aligns with data protection regulations.

✅ Ease of Use: Simplifies PII protection with pre-built and customizable tools.

By leveraging PII Masking, developers can create secure and privacy-compliant AI workflows without compromising functionality or user experience.

What is a Sub-Agent?

A Sub-Agent is a specialized component within an AI Agent that handles a specific set of tasks related to a particular domain or functionality. It acts as a modular processing unit, receiving tasks from the main agent after identifying user intent.

Think of Sub-Agents as modules in a typical application—they focus on distinct functionalities within the AI system.

How Sub-Agents Work

1. User Query Processing

   • The main agent receives the user’s request and determines the intent.

   • If the intent falls within a specific domain, it is delegated to the appropriate Sub-Agent.

2. Task Execution by Sub-Agent

   • The Sub-Agent processes the request based on the identified user intent.

   • It may extract information, analyze data, summarize content, or answer user queries by delegating the execution to right intent.

3. Response Generation

   • The Sub-Agent compiles the required information by executing the workflow associated with the intent and sends the response back to the main agent.

   • The main agent then formats and delivers the final response to the user.

Example: Document AI Agent

A Document AI Agent may have a Document Q&A Sub-Agent, which:

• Handles user queries about uploaded documents.

• Retrieves, extracts, and summarizes information from documents like paystubs, invoices, and balance sheets.

• Ensures users get accurate responses based on the document data processed by the AI.

How to create sub-agent?

See the video below to learn how to create Sub-Agents:

Key takeaway for developers

Why Use Sub-Agents?

✅ Modular Structure: Organizes AI workflows efficiently.

✅ Scalability: Allows for adding specialized capabilities without modifying the entire agent.

✅ Improved Accuracy: Sub-agents focus on specific tasks, enhancing performance.

By using Sub-Agents, developers can build structured, scalable AI systems that efficiently handle complex, multi-domain tasks.

Overview

The Data Gateway feature in UPTIQ AI enables developers to seamlessly connect their AI agents to a wide array of third-party services. By configuring authentication parameters such as API keys and OAuth credentials, developers can access both raw and normalized data through the Data Gateway's APIs, enhancing the capabilities of their AI-driven applications.

Overview

In UPTIQ Workbench, Nodes are the fundamental building blocks of workflows, enabling developers to define, configure, and execute business processes in a declarative manner. Nodes represent distinct actions or tasks within a workflow, allowing seamless automation of AI-driven operations.

A Node in UPTIQ follows the traditional concept of nodes in Business Process Automation (BPA) systems, where each node is responsible for executing a specific function—such as processing data, making decisions, fetching external resources, or displaying information to users.

Nodes help structure complex workflows by: ✅ Modularizing logic into discrete, reusable components. ✅ Enabling automation by orchestrating tasks in a predefined sequence. ✅ Reducing code dependency by offering configuration-driven business logic.

In essence, Nodes allow developers to build scalable, structured, and AI-enhanced workflows without writing extensive custom code.

Categories of Nodes in UPTIQ

Nodes in UPTIQ are divided into five major categories, each catering to a different aspect of workflow automation.

1. Data Nodes (Database Interaction)

These nodes provide powerful database interaction capabilities to store, retrieve, and query data efficiently. Developers can use Data Nodes to: ✅ Write, read, or query databases, including structured and unstructured data. ✅ Integrate with external databases to fetch real-time information. ✅ Store and retrieve AI-processed data for future queries.

Example Use Cases:

• Storing extracted financial data from invoices in a structured table.

• Fetching historical transactions for user queries.

2. Integration Nodes (Third-Party & External System Interaction)

These nodes allow developers to connect workflows with external systems. UPTIQ provides Data Gateway integration, but developers can also connect directly to third-party services.

✅ Fetch data from external CRMs, ERPs, or financial platforms. ✅ Send data to third-party systems via APIs. ✅ Automate multi-system workflows by integrating AI agents with business applications.

Example Use Cases:

• Fetching vendor details from an accounting system when processing an invoice.

• Syncing loan application status with an external credit-checking API.

3. User Interaction Nodes (User Engagement & Frontend Interactions)

These nodes help developers interact with users within a workflow by collecting inputs, showing outputs, or guiding users with messages.

✅ Capture user input dynamically (e.g., text, dropdown selection). ✅ Display messages, confirmation prompts, and loaders. ✅ Present structured data using tables, links, or charts.

Example Use Cases:

• Asking a user for approval before proceeding with document processing.

• Showing a clickable link to download a financial report.

4. AI Foundational Nodes (GenAI & AI Capabilities)

One of the most critical node categories in UPTIQ, AI Foundational Nodes unlock Generative AI (GenAI) and advanced AI functionalities, enabling AI-driven automation.

✅ Enable Retrieval-Augmented Generation (RAG) to enhance AI responses with real-time data. ✅ Access Large Language Models (LLMs) for content generation, intent classification, and response optimization. ✅ Integrate AI-driven web crawling to fetch and process external knowledge.

Example Use Cases:

• Using LLM nodes to summarize financial contract documents.

• Implementing RAG nodes to enhance AI responses with proprietary business knowledge.

• Classifying user queries with an Intent Classification Node.

5. Standard Nodes (Workflow Orchestration & Control)

This category contains miscellaneous utility nodes that help in workflow orchestration, modularization, and process control. These nodes ensure the smooth execution of workflows by handling logical conditions, loops, and modular workflow structures.

✅ Implement conditional branching, loops, and state transitions. ✅ Build reusable workflow modules for better maintainability. ✅ Control the execution sequence within complex workflows.

Example Use Cases:

• Routing users through different processes based on document type.

• Modularizing repetitive workflows like data extraction and validation.

Key Takeaways for Developers

✅ Node is fundamental building block to help developers implement business logic in declarative manner. ✅ Each node category has a specific purpose—data handling, integration, user interaction, AI processing, or workflow control. ✅ AI Foundational Nodes enable advanced AI-driven automation and GenAI capabilities. ✅ Integration Nodes connect workflows with external systems for real-time data exchange. ✅ User Interaction Nodes improve engagement by presenting structured outputs and collecting inputs. ✅ Standard Nodes help with logic control, modularization, and process optimization.

By understanding Nodes in UPTIQ, developers can design highly flexible, efficient, and AI-powered workflows that automate complex business processes with minimal effort.

Welcome to the exciting world of Generative AI! In this section, we’ll explore some cool and essential ideas for developers itching to create amazing AI agents. We’ll dive into topics like understanding user intents, making inferences, leveraging large language models, and automating AI workflows.

Get ready to tackle tough challenges and discover awesome possibilities to build smart, scalable AI solutions!

Overview

The Loop Node in UPTIQ Workbench allows workflows to repeatedly execute a process until a specified exit condition is met. This is particularly useful for iterating over lists, verifying input validity, or ensuring a task completes before proceeding.

By using the Loop Node, developers can: ✅ Handle Iterative Processing – Process lists of items dynamically. ✅ Validate User Input Repeatedly – Ensure correct data entry before proceeding. ✅ Create Conditional Workflows – Execute repeated steps based on runtime conditions.

Unlike static workflows, the Loop Node dynamically determines whether to continue execution or exit based on real-time data evaluation.

Configurations

Execution Flow:

1️⃣ The Loop Node receives input from previous steps. 2️⃣ It evaluates the exit condition based on the provided data. 3️⃣ If the condition is met, the node outputs {"action": "exit"} and the loop terminates. 4️⃣ If the condition is NOT met, it outputs {"action": "continue"}, repeating the process.

Output Format

Example Use-Cases

Use-Case 1: Validating User Email Input

A workflow requires users to enter a valid email before proceeding. The Loop Node repeats the request until a valid email format is entered.

Configuration:

Execution Process:

• If the user enters an invalid email, the node outputs:

  🔹 The workflow asks the user to re-enter their email.

• Once the user enters a valid email, the node outputs:

  🔹 The workflow proceeds to the next step.

Use-Case 2: Iterating Through a List of Questions

A chatbot is programmed to ask multiple questions, and the workflow loops until all questions are answered.

Configuration:

Input to Loop Node:

Execution Process:

1️⃣ First Iteration → The bot asks Q1, removing it from the list.

2️⃣ Second Iteration → The bot asks Q2, removing it from the list.

3️⃣ Third Iteration → The bot asks Q3, removing it from the list.

🔹 Once no questions remain, the workflow exits the loop.

Use-Case 3: Retry Logic for API Requests

A workflow fetches data from an external API. If the API fails, the Loop Node retries the request until a successful response is received or retries are exhausted.

Configuration:

Execution Process:

1️⃣ If the API fails, the node outputs:

🔹 The API call is retried.

2️⃣ If the API succeeds or retries reach 3, the node outputs:

🔹 The workflow proceeds to handle the response or logs an error.

Key Takeaways for Developers

✅ Automates Loop Execution – Runs a process repeatedly until a condition is met.

✅ Uses Natural Language for Exit Conditions – Unlike traditional programming logic, the Loop Node allows developers to define exit conditions in plain language, making it easy to configure and readable within workflows.

✅ Reduces Redundant Workflow Steps – Instead of creating multiple nodes for repetitive tasks, use a Loop Node for dynamic iteration.

✅ Enables Smart Decision-Making – The exit condition is evaluated dynamically, ensuring real-time logic execution.

✅ Supports Multiple Use Cases – Ideal for validations, API retries, question sequences, and dynamic task execution.

By integrating the Loop Node, developers can create intelligent, adaptable workflows that respond dynamically to user input, process execution results, and iterative data operations. 🚀

What is Knowledge in Agent?

In the context of AI agents on UPTIQ, Knowledge refers to the repository of structured and unstructured information that the agent relies on to better understand user queries and provide accurate responses. This knowledge is typically stored in a RAG (Retrieval-Augmented Generation) container, which combines traditional information retrieval techniques with generative AI to enhance the agent's reasoning and intent classification capabilities.

• RAG Container: A hybrid system where relevant data is retrieved from the knowledge base and passed to the reasoning engine for context-aware responses.

• Role in Reasoning: Knowledge is not just static data but a dynamic resource that informs the Reasoning Engine when interpreting or classifying user intents. It allows the agent to understand nuanced queries, connect them with relevant information, and generate meaningful responses.

How Does it Work?

1. Data Storage:

   • Knowledge is preloaded into the RAG container.

   • It may consist of documents, FAQs, databases, product manuals, or any domain-specific resources.

   • Information can be structured (e.g., tabular data) or unstructured (e.g., natural language text).

Examples

1. Customer Support Agent:

   • Knowledge: A collection of FAQs, product manuals, and troubleshooting guides.

   • Functionality: When a customer asks about a specific product feature, the RAG container retrieves the relevant section of the manual, which the reasoning engine uses to generate a response.

Key takeaway for developers

In UPTIQ, Knowledge serves as the foundation for intelligent reasoning in AI agents. It equips agents with the ability to:

✅ Contextualize user queries.

✅ Refine intent classification.

✅ Provide factually correct and relevant responses.

Developers should focus on curating high-quality, domain-specific knowledge bases to maximize the accuracy and utility of their AI agents.

Overview

The Web Crawler Node in UPTIQ Workbench is designed to extract relevant information from web pages in real-time. Unlike traditional web scrapers, this node is optimized for AI-driven workflows, where extracted content can be processed by Large Language Models (LLMs) to generate structured insights.

This node is particularly useful for retrieving dynamic, publicly available information, such as company overviews, industry trends, or competitor insights. The extracted data can be refined, summarized, and structured to fit business needs, making it a valuable component for AI-driven research and automation.

Configurations

URL (Required)

• The fully-qualified web address from which data should be retrieved.

• Example: https://www.uptiq.ai/about

Instructions (Required)

• Defines how the extracted web content should be processed.

• Instructs AI agents on what aspects of the data to analyze and summarize.

• Example:

Output Format

• The Web Crawler Node outputs structured data in JSON format.

• Example output:

Example Use-Case

1. Summarizing a Company's Information

Scenario: A user requests an overview of a company. The Web Crawler Node scrapes the company's "About" page and passes the extracted content to an LLM node, which generates a concise, structured summary.

Workflow Nodes Used in this Use-case

1. Web Crawler - to scrape and summurize the information

2. Display - to display the information to user.

Configurations:

Workflow Steps:

1. Web Crawler Node scrapes https://www.uptiq.ai/about to extract relevant content.

2. LLM Node processes the extracted content and generates a structured summary.

3. Display Node presents the final output to the user.

Final Output:

Key Takeaways for Developers

✅ Real-Time Data Extraction – The Web Crawler Node retrieves fresh, publicly available content from websites for AI processing.

✅ Structured AI-Driven Summarization – Extracted content is refined using LLMs, ensuring concise and contextually relevant outputs.

✅ Customizable Processing Instructions – Developers can tailor how extracted data is interpreted and structured by modifying the instructions field.

✅ JSON-Formatted Output – Ensures compatibility with other workflow components for seamless data handling.

By integrating the Web Crawler Node into workflows, developers can automate web-based data retrieval and AI-powered summarization, significantly enhancing information accessibility and decision-making. 🚀

The Upload Document Node in UPTIQ Workbench provides a seamless way to ingest documents into a workflow, enabling users to extract information, process files, and interact with document content dynamically.

This node supports multiple upload methods, allowing developers to: ✅ Accept file uploads directly from users. ✅ Retrieve documents via signed URLs from external systems. ✅ Process base64-encoded documents for advanced automation.

Once a document is uploaded, the node returns a documentId, which serves as a reference for further processing—such as: 🔹 Extracting text and data from documents. 🔹 Converting documents into images or zipped archives. 🔹 Passing the document to the Prompt Node for AI-driven queries. 🔹 Uploading the file to an external storage system via pre-signed URLs.

The Upload Document Node is a core component in document-driven workflows, making it easy to process files dynamically and integrate them with

Overview

The Table Write Node in UPTIQ AI Workbench allows developers to store, update, and manage structured data within an agent’s persistent storage layer. Unlike traditional databases, UPTIQ’s Table concept provides a simplified yet effective way to maintain structured data that remains accessible across workflows.

This node is crucial for workflows requiring data persistence, such as tracking transactions, maintaining user records, logging workflow actions, and managing application statuses.

Generative AI Models

A Generative AI Model is an advanced artificial intelligence (AI) model designed to process and generate human-like text based on vast amounts of data. It is trained using deep learning techniques, particularly transformer architectures (like GPT, BERT, or LLaMA), and can understand, predict, and generate language in a way that mimics human communication.

What It Is?

Generative AI models are algorithms that generate new content based on patterns they have learned from data. Instead of just analyzing or classifying data, they create text, images, music, or even code.

How Does It Work?

Think of it like a student learning to write essays by reading thousands of articles. Over time, the student can write original essays that sound natural. Generative AI does the same but much faster.

Examples:

• GPT-4 (by OpenAI): Writes text, answers questions, helps with coding.

• DALL·E 3: Creates images from text descriptions.

• Stable Diffusion: Generates art and graphics based on inputs.

Where It’s Used?

• Writing emails, blogs, and reports.

• Generating financial summaries.

• Creating marketing images and designs.

Overview

The JavaScript Node in UPTIQ Workbench enables developers to execute custom JavaScript code within a workflow. While built-in workflow nodes handle many automation tasks, there are scenarios where custom logic, data transformation, or conditional operations are required. The JavaScript Node provides the flexibility to manipulate, filter, or format data dynamically before passing it to the next step in the workflow.

With this node, developers can: ✅ Perform data transformation by modifying, formatting, or restructuring JSON objects. ✅ Execute mathematical computations such as tax calculations or discount applications. ✅ Implement conditional logic to alter workflow paths based on input values. ✅ Merge, clean, or reformat API responses before passing them to the next node. ✅ Work with context variables, agent-level data, and secret variables for secure and dynamic processing.

Configurations

Execution Flow

1️⃣ The JavaScript Node receives input from previous workflow steps. 2️⃣ It executes custom JavaScript logic, transforming or processing the input. 3️⃣ The output of the script is passed to the next node in the workflow.

Example Syntax for Variable Usage

Example Use-Cases

Use-Case 1: Formatting API Response Data

A workflow fetches user details from an external API, but the response includes unnecessary fields. The JavaScript Node is used to extract and reformat the required fields while also deriving a new field (location).

API Response (From Previous Node)

JavaScript Node Code Snippet

Output Passed to Next Node

🔹 How It Helps: ✔ Removes unnecessary fields. ✔ Combines multiple fields into a structured response. ✔ Prepares the data for the next workflow step.

Use-Case 2: Applying Conditional Business Logic

A loan eligibility check requires that if a user's credit score is below 650, a flag should be set for manual review.

JavaScript Node Code Snippet

Output Passed to Next Node

🔹 How It Helps: ✔ Automates eligibility checks without requiring a separate ruleset. ✔ Streamlines manual review processes based on conditions.

Use-Case 3: Calculating Discounts Based on Order Total

A workflow calculates a discount percentage based on an order’s total amount.

JavaScript Node Code Snippet

Output Passed to Next Node

🔹 How It Helps: ✔ Implements dynamic business logic for discount calculation. ✔ Reduces dependency on external services for simple calculations.

Best Practices & Key Takeaways for Developers

✅ Always Return an Output – Ensure the script returns a valid JavaScript object or primitive value, as this is passed to the next node.

✅ Validate Inputs – Use checks to avoid undefined values or workflow failures.

✅ Use Secret Variables Securely – When working with API keys or sensitive data, store them in secret.<var_name> instead of hardcoding values.

✅ Optimize Performance – Keep scripts lightweight to avoid workflow execution delays.

✅ Error Handling is Essential – Use try...catch to gracefully handle failures within the script.

Example Error Handling Pattern

By integrating the JavaScript Node, developers can extend workflow functionality beyond built-in nodes, allowing for custom logic execution, data transformation, and dynamic workflow adaptability. 🚀

Overview

Rulesets in UPTIQ enable developers to define business rules that can be executed within agentic workflows to automate decision-making. These rules help streamline processes like loan origination, compliance validation, and data filtering based on predefined conditions.

For example, in a loan origination workflow, a ruleset can be used to automatically filter out loan applications where:

• The loan amount is less than $1000

• The borrower’s age is under 18

By defining these conditions in a Ruleset, the AI agent can evaluate applications instantly and proceed only with the ones that meet the eligibility criteria.

Creating a Ruleset in UPTIQ

1. Navigate to Config & Utils → Click "Create Ruleset"

2. Enter a Name for the Ruleset and Save.

3. Define Facts (Input Variables):

   • Click on the created Ruleset and select "Create Fact."

Using a Ruleset in Workflows

Once a Ruleset is created, it can be integrated into a workflow to dynamically evaluate conditions and automate decision-making. Let’s go step by step with an example.

Example Workflow Scenario

Suppose the previous node in the workflow is a JS Node that processes loan application data and produces the following JSON output:

In UPTIQ workflows, each node’s output serves as the input for the next node. This means the Ruleset Node will receive the above JSON data as input.

Mapping Runtime Values to Ruleset Facts

To enable the Ruleset to evaluate conditions dynamically, we map input variables from the previous node’s output to the corresponding Facts defined in the Ruleset.

1. Drag a "Ruleset Node" into the workflow

   • Position it after the JS Node that generates the loan application data.

2. Select the Created Ruleset

Execution & Decision Evaluation

• The Ruleset Node evaluates the conditions:

  • ✅ Loan Amount (1500) is greater than $1000 → Rule Passes

  • ✅ Borrower Age (25) is greater than 18 → Rule Passes

• If all required conditions are met, the output variables

What Happens Next?

• If both rules pass, the workflow proceeds to the next step, such as approval, document processing, or further validations.

• If one or more rules fail, the output variable for that rule will not be included, allowing developers to implement alternative paths, such as rejection or additional review.

Key Takeaways for Developers

✅ Automate Business Logic – Use Rulesets to define and execute structured decision-making processes without manual intervention. ✅ Flexible & Scalable – Define multiple rules within a Ruleset to support complex decision-making in AI workflows. ✅ Real-Time Rule Execution – Pass dynamic runtime values to evaluate conditions in real time. ✅ Seamless Workflow Integration – Easily integrate Ruleset nodes in workflows to automate approvals, filter data, or trigger actions based on rule outcomes.

By leveraging Rulesets, developers can create smarter AI agents that make automated, context-aware decisions, improving efficiency and accuracy across various AI-driven processes.

Overview

The Document to Image Node abstracts the process of converting any document (PDF, DOCX, Excel, etc.) into images, making it easier to pass structured data to LLMs for extraction.

This node is critical in workflows where: ✅ Documents need to be summarized before being processed. ✅ Text extraction accuracy needs improvement (reducing formatting errors). ✅ OCR and AI-driven tools require images for better text recognition. ✅ Structured data from Excel sheets needs to be extracted accurately (e.g., financial tables).

What is a workflow ?

Understanding UPTIQ AI Workbench Workflows

Workflows in UPTIQ AI Workbench represent low-code to no-code solutions designed to help developers implement business logic for specific AI agents. These workflows act as the

Overview

The Storage feature in UPTIQ Workbench functions similarly to Tables but is specifically designed for document storage. While Tables handle structured data storage, Storage allows AI agents to store, retrieve, and manage documents across workflows. This ensures that AI-driven processes have seamless access to necessary files for decision-making, document processing, and data extraction.

With Storage, developers can: ✅ Store various document types (e.g., PDFs, images, scanned forms) that AI agents process. ✅ Retrieve stored documents dynamically within workflows. ✅ Enable AI models to reference documents

Developer Support Information

For any development-related queries, assistance, data gateway integrations, or troubleshooting while working with the Uptiq AI Workbench, developers can reach out to our support team at . Whether you need help with agent configuration, integrations, workflows, or debugging, our team is available to provide guidance and resolve issues efficiently.

Feel free to contact us for:

• Technical support on building and deploying AI agents.

RAG, or Data Retrieval & Augmentation, is like giving your AI model a superpower to find and use extra information when it needs it.

Why is it needed?

AI models are only as good as the data they're trained on. Sometimes, that data might not be enough to answer a question or complete a task accurately. RAG solves this problem by letting the AI model access and use additional, relevant information from external sources.

Get started with an UPTIQ AI agent. Your own AI Assistant.

While LLMs are powerful, their capabilities are bound by their pre-existing knowledge. With UPTIQ AI Agent, you can go beyond these limitations by integrating customized, relevant workflows to supercharge your AI agent. Whether it’s leveraging prebuilt workflows or designing your own, the Workbench empowers you and your team to create AI solutions tailored to your needs. Your AI assistant is now smarter, faster, and more aligned with your goals.

What does it do?

The agent is the main point of contact for users, handling their questions and getting things done. They're the decision-maker, figuring out what needs to be done to meet user requests and the best way to do it. This key role makes the agent the center of user interaction and task completion, highlighting their importance in ensuring smooth user experiences and efficient results.

What is Intent Classification?

Intent classification is the process of identifying the purpose or goal behind a user’s input in an AI-driven application. It enables AI agents to determine what the user wants and route the request to the correct workflow or response.

Importance of Intent Classification

• Helps AI applications understand user queries accurately.

• Routes the user to the correct process or action.

As a developer, why should you know about AI Agents?

Understanding AI agents is crucial for developers as they are central to building intelligent and autonomous systems. By mastering the design, development, and deployment of AI agents, developers can unlock a new era of innovation, streamlining complex processes and enhancing user interactions. These agents can be programmed to learn and adapt, making them invaluable for tasks that require decision-making and problem-solving skills. Furthermore, AI agents can operate autonomously, reducing the need for human intervention and increasing efficiency.

Developers proficient in AI agent frameworks and tools can accelerate development cycles and remain at the forefront of the rapidly evolving AI landscape. These frameworks provide a foundation for building intelligent agents, while tools facilitate tasks such as data collection, model training, and agent deployment. By leveraging these resources, developers can create sophisticated AI agents capable of tackling a wide range of challenges.

Moreover, AI agents can be integrated into various applications, from customer service chatbots to autonomous vehicles. This versatility makes them an essential tool for developers across industries. As AI technology continues to advance, we can expect even more innovative applications of AI agents, further solidifying their importance in the field of software development.

What is Generative AI?

Generative AI refers to artificial intelligence systems that can create new content, such as text, images, music, and even code, by learning from patterns in data. Unlike traditional AI models that classify or predict based on existing data, generative AI can generate novel outputs.

Overview

The Entity Recognizer (ER) feature enables developers to define custom patterns to identify specific entities in natural language user queries. These patterns are used to extract relevant entities (e.g., email addresses, loan application numbers, phone numbers) that are critical for processing workflows efficiently.

For example: A user submits the query, "Give me the amortization schedule for my loan application: LA:12312331."

• Developers can define a custom regex to recognize the loan application number LA:12312331.

• The Entity Recognition Node in the workflow can then use this pattern to extract the entity dynamically from the user query.

How to Create a New Entity Recognizer

1. Navigate to the Model Hub → Select the Entity Recognizer tab.

2. Click "Create Entity Recognizer."

3. Fill in the following fields:

   • Name: Provide a meaningful name (e.g., Loan Application Recognizer

Using an Entity Recognizer in Workflows

To use a created Entity Recognizer, follow these steps:

1. Add the Entity Recognition Node to the workflow.

2. Pass the user query as input to the node.

   • Use the special variable $agent.query, which contains the user’s natural language query.

Example:

• Entity Recognizer: Loan Application Number

• Input: Give me the amortization schedule for my loan application: LA:12312331

• Output:

The output from this node will be used as input for the next connected node, where the named entity can be accessed using the 'result' key

Key Use Cases for Entity Recognizers

✅ Extract Named Entities – Recognize and extract structured data (e.g., email addresses, IDs, dates) from user queries. ✅ Enhance Workflow Automation – Use extracted entities to dynamically route workflows or fetch related data. ✅ Handle Complex Inputs – Process unstructured natural language queries with precision.

Key Takeaways for Developers

✅ Customizable Patterns – Define regex patterns to recognize specific entities tailored to your use case. ✅ Seamless Workflow Integration – Use the Entity Recognition Node to incorporate entity extraction directly into workflows. ✅ Efficient Processing – Simplify user query handling by dynamically identifying and isolating critical information. ✅ Regex Flexibility with Flags – Use flags to adjust pattern behavior for better matching and adaptability.

By leveraging Entity Recognizers, developers can create AI workflows that intelligently extract and process key information, improving the efficiency and accuracy of automated responses.

In today's fast-paced digital landscape, where efficiency and accuracy are paramount, AI workflow automation has emerged as a transformative force. It's not merely a technological advancement; it's a strategic imperative that empowers businesses to optimize operations, enhance productivity, and unlock new realms of innovation.

Why is AI Workflow Automation Important?

• Efficiency and Productivity: By automating repetitive and mundane tasks, AI liberates human workers to focus on strategic, creative, and value-added activities. This streamlines processes, reduces errors, and accelerates turnaround times, leading to enhanced productivity and operational efficiency.

• Cost Savings: Automation reduces the need for manual labor, leading to significant cost savings in the long run. Additionally, by minimizing errors and optimizing resource allocation, AI workflow automation helps businesses avoid costly rework and delays.

• Scalability: AI-powered workflows can be easily scaled to accommodate growing business needs. This flexibility enables organizations to adapt to changing market conditions and seize new opportunities without incurring significant additional costs.

• Data-Driven Insights: AI workflow automation generates a wealth of data that can be leveraged to gain valuable insights into business operations. These insights can be used to identify bottlenecks, optimize processes, and make informed decisions.

• Improved Customer Experience: By automating customer-facing tasks such as order processing and support, AI can deliver faster, more personalized, and more consistent customer experiences. This can lead to increased customer satisfaction and loyalty.

• Innovation and Growth: By freeing up resources and enabling faster, more efficient operations, AI workflow automation fosters a culture of innovation. This empowers businesses to explore new ideas, develop new products and services, and stay ahead of the competition.

In essence, AI workflow automation is not just about doing things faster; it's about doing things smarter. It's about leveraging the power of artificial intelligence to transform the way businesses operate, compete, and grow in the digital age.

How AI Workflow Automation Enhances AI Agents

1. Efficiency and Focus:

   • AI workflow automation handles repetitive tasks, allowing the AI agent to concentrate on higher-level functions like natural language understanding and decision-making.

   • This division of labor improves the overall efficiency and effectiveness of the AI agent.

2. Scalability and Adaptability:

Key Takeaway for Developers:

By incorporating AI workflow automation into the design and development of AI agents, you can create more intelligent, efficient, and adaptable systems that deliver superior results. Remember that the AI agent is the "brain" that makes decisions and takes action, while the AI workflow automation is the "backbone" that supports and enhances its capabilities.

Overview

The Input Node in UPTIQ Workbench is designed to collect user input dynamically within a workflow, ensuring that processes requiring user-provided data can proceed efficiently. This node plays a crucial role in workflows where structured or freeform input is required before the next action is executed.

Unlike static configurations, the Input Node enables real-time user interaction, allowing workflows to adapt based on user-provided values. It supports multiple input types, including basic text, numbers, and rich text formatting, making it ideal for structured and detailed data collection scenarios.

Retrieval-Augmented Generation (RAG) is a framework that combines traditional information retrieval with generative AI. It enables AI agents to generate contextually accurate and factually grounded responses by retrieving relevant information from a knowledge base or data source and using it to augment the reasoning process.

In UPTIQ AI Workbench, RAG is implemented using a sequence of components designed to manage and use data effectively in AI workflows. Below is a detailed explanation of each related concept:

How to create a RAG?

Large Language Models (LLMs)

UPTIQ provides access to foundational LLMs from leading providers such as OpenAI, Meta, Google, Anthropic, and Groq. This allows developers to experiment with different models and evaluate their behavior for specific AI use cases within their agents.

Exploring Different Model Capabilities

Each LLM has unique strengths that developers can leverage based on their needs:

Overview

The Loader Node in UPTIQ Workbench is designed to enhance user experience by displaying a processing message when a workflow involves a time-consuming operation.

When workflows require retrieving external data, running computations, or executing multi-step processes, the Loader Node ensures that users are aware of ongoing activity rather than experiencing delays with no feedback.

This helps to reduce user frustration, improve engagement, and create a smoother conversational flow by keeping users informed during system processing.

What is Inference?

Inference is the process of running a trained AI model on new data to generate predictions or insights. It is the execution phase where an AI system applies learned knowledge to new situations.

For Example:

Imagine you're building a task management app that helps users prioritize their to-do list. You’ve integrated an AI feature that analyzes tasks and suggests priorities (e.g., "High," "Medium," or "Low") based on past behavior. When a user adds a new task, such as "Prepare quarterly report," the app runs it through a pre-trained AI model. The model analyzes the task's description and matches it to patterns learned from past tasks (like similar descriptions being labeled as "High Priority"). Based on this, the model suggests: "High Priority".

This is inference in action—using a trained model to make decisions or predictions for new, unseen data.

Importance of Inference

Large Language Models (LLMs) are a type of Generative AI model that focuses on understanding and generating human-like text. They are trained on vast amounts of text data and can write, summarize, translate, and even code based on input prompts.

How It Works?

• The model analyzes billions of words from books, articles, and the internet to learn language structure.

• When given a prompt, it predicts the most likely next words based on its training.

Overview

Tables in Workbench provide an abstract database layer that allows developers to persist data within agentic workflows. Think of it as a custom data storage system where developers can define and manage structured records without worrying about underlying database complexities.

This feature is particularly useful when AI agents need to store, retrieve, and manipulate data across workflows—such as tracking user interactions, logging processed data, or maintaining reference records.

🔹 Storage Limitation: Persistent database space is limited by default but can be expanded separately if developers anticipate storing large volumes of data within AI agents.

How to Create a Table?

There are two ways to create a Table in UPTIQ:

1. From Config & Utils (Recommended for Reusability)

1. Navigate to Config & Utils → Tables Tab.

2. Click "Create Table."

3. Enter a name for the table (this serves as a unique identifier for data storage).

4. Click Save—the table is now ready for use in workflows.

2. Directly from a Workflow Node (Quick Method)

1. Drag a Table Write/Read node into a workflow.

2. Click on the node to open the side panel.

3. Click "Add Table"—this automatically assigns the table to the current agent.

4. Save the table definition.

How to Use Tables in AI Agent Development

Developers can leverage persistent tables to:

✅ Store Data Across Workflow Executions – Maintain records of AI-generated outputs, processed user inputs, or any other structured data.

✅ Retrieve & Reuse Data – Query stored information to enhance AI responses, track user history, or fetch reference data dynamically.

✅ Automate Business Logic – Use tables to store intermediate results that can be accessed by multiple workflow nodes, reducing redundant computations.

✅ Enable Long-Term Data Persistence – Unlike temporary workflow variables, tables retain data across workflow executions, allowing AI agents to operate with stateful memory.

Best Practices for Using Tables in Workflows

✔ Design Tables Thoughtfully – Only store data that is required for AI workflows. Avoid persisting unnecessary information to conserve storage.

✔ Regularly Clean Up Data – Since storage space is limited, implement periodic clean-ups for expired or unnecessary records.

✔ Ensure Data Security & Compliance – Be mindful of storing sensitive user data and implement access control mechanisms where required.

Key Takeaways for Developers

✅ Use Tables to Persist AI Data – Maintain structured records that workflows can access and update dynamically. ✅ Store & Query Information Efficiently – Design data models that support your AI agent’s functionality. ✅ Leverage Tables for Stateful AI Workflows – Enable AI agents to retain context between workflow executions. ✅ Manage Storage Effectively – Be mindful of storage limitations and optimize data persistence strategies.

By integrating Tables in workflows, developers can enhance AI capabilities, improve decision-making, and enable long-term data-driven automation.

Overview

Like many low-code/no-code platforms, UPTIQ allows developers to inject custom code into workflows when required. The Workbench provides the ability to execute JavaScript code using Code Snippets, which can be embedded in workflows through the "Javascript" Node.

What is a code snippet?

A Code Snippet is a simple JavaScript function named main, which:

• Takes input from the previous node in the workflow.

• Stores that input in a special variable input, making it available throughout the function.

• Outputs the processed result, which becomes the input for the next node in the workflow.

Understanding 'Input' in Code Snippets

Since every node’s output becomes the input for the next node, the data type of the input variable depends on the preceding node’s output:

• If the previous node outputs a JSON object, input will be of type JSON.

• If the previous node outputs a string or number, input will hold that respective type.

This flexibility allows developers to manipulate, transform, or filter data dynamically within workflows.

When to Use Code Snippets?

Developers can use JavaScript Nodes for: ✅ Data Formatting & Transformation – Convert data formats, adjust values, or prepare data for API calls. ✅ Basic Filtering – Remove unnecessary fields or extract key data before passing it to the next node. ✅ Applying Business Logic – Execute programmatic operations that aren’t natively supported by other workflow nodes. ✅ Custom Computations – Perform calculations or data aggregations before the AI agent processes the result.

Best Practices & Pitfalls of Using JavaScript Nodes

🚨 Avoid Overuse of JavaScript Nodes While JavaScript nodes add flexibility, excessive use can:

• Reduce Workflow Readability – Business logic may become hidden inside code, making workflows harder to understand.

• Complicate Debugging & Maintenance – Unlike visual nodes, JavaScript logic isn’t easily visible, which can make troubleshooting time-consuming.

• Break the Low-Code Advantage – Workflows should remain as low-code as possible; use JavaScript only when absolutely necessary.

💡 Recommendation: Use JavaScript Nodes sparingly. Always check if the same transformation or filtering can be achieved with other workflow nodes before resorting to custom code.

How to create a code snippet?

There are two ways to create a Code Snippet:

1. From Config & Utils

1. Navigate to Config & Utils → Code Snippet Tab.

2. Click "Create Code Snippet."

3. Enter a meaningful name for easy identification in workflows.

4. Select the agent where the code snippet will be used.

2. Directly from a JavaScript Node (Quick Method)

1. Drag a JavaScript Node into the workflow.

2. Click on the node to open the side panel.

3. Click "Add Script"—this creates a new snippet and automatically assigns it to the agent.

4. Save the snippet and later update it with the required business logic.

Key Takeaways for Developers

✅ Leverage JavaScript Nodes for targeted logic execution, such as data transformations and filtering. ✅ Ensure workflow readability by limiting JavaScript usage to essential cases. ✅ Use reusable Code Snippets to maintain modularity and avoid redundant script writing. ✅ Prioritize low-code workflow nodes whenever possible for better maintainability and collaboration.

By using Code Snippets effectively, developers can enhance workflow capabilities without compromising the benefits of low-code/no-code development in UPTIQ.

Overview

The API Call Node in UPTIQ Workbench enables workflows to interact with external APIs by performing HTTP requests. This node allows developers to fetch, send, update, or delete data from external services in a structured and automated manner.

By leveraging the API Call Node, developers can integrate their workflows with third-party APIs, internal services, or cloud endpoints, ensuring seamless data exchange between systems.

Configurations

Endpoint (Required)

• The URL of the external API to which the request is sent.

• Supports static URLs or dynamic variables fetched from workflow data.

Method (Required)

• Defines the type of HTTP request to be made.

• Supported methods: ✅ GET – Retrieve data from the API. ✅ POST – Send new data to the API. ✅ PUT – Update an existing resource. ✅ PATCH – Modify part of an existing resource. ✅ DELETE – Remove a resource.

Headers (Optional)

• A set of key-value pairs representing HTTP headers to include in the request.

• Example: { "Authorization": "Bearer <token>", "Content-Type": "application/json" }.

Parameters (Optional)

• Query parameters to be appended to the URL.

• Example: { "userId": "12345" } results in https://api.example.com/resource?userId=12345.

Request Body (For POST, PUT, PATCH requests only)

• The payload sent with the request.

• Can be in raw JSON format or key-value format.

• Example: { "name": "John Doe", "email": "[email protected]" }.

Output Format

• If the request is successful, the node outputs:

• If the request fails, the node outputs:

Example Use-Cases

Example 1: Fetching Data from an External API (GET Request)

A workflow needs to retrieve user details from a third-party service.

Configuration:

• Endpoint: https://jsonplaceholder.typicode.com/todos/1

• Method: GET

Response Output:

Example 2: Sending Data to an API (POST Request)

A workflow needs to create a new user record in an external system.

Configuration:

• Endpoint: https://api.example.com/users

• Method: POST

• Headers: { "Content-Type": "application/json" }

Response Output (Success):

Response Output (Error):

Key Takeaways for Developers

✅ Seamless API Integration – Easily connect UPTIQ workflows with external APIs for data exchange. ✅ Supports All Major HTTP Methods – Perform GET, POST, PUT, PATCH, and DELETE requests. ✅ Flexible Configuration – Customize headers, query parameters, and request body as per API requirements. ✅ Handles API Responses Efficiently – Captures both successful data responses and error messages for better debugging. ✅ Use with Dynamic Variables – Fetch endpoint URLs and request data dynamically from previous nodes for dynamic API calls.

By leveraging the API Call Node, developers can extend UPTIQ Workbench workflows beyond internal processes, integrating them with external platforms, databases, and third-party applications for automated and intelligent data handling. 🚀

Overview

The Ruleset Node in UPTIQ Workbench serves as an abstraction that allows developers to integrate Rulesets into workflows seamlessly. It helps evaluate specific records against predefined business rules.

Rulesets are defined separately and can be used to enforce business logic, such as eligibility checks, risk assessments, or compliance validation. The Ruleset Node enables developers to apply these rules dynamically to real-time data, ensuring that decisions are made based on runtime inputs.

For detailed information on how Rulesets are created and managed, refer to the Ruleset Section in the developer guide.

Configurations

Mappings Explained

Once a Ruleset is selected, developers must map input data to the Facts defined in the Ruleset. This mapping consists of:

1. Target (Fact in Ruleset)

   • This field is auto-populated based on the Facts defined in the selected Ruleset.

   • Example: If the Ruleset has Facts LoanAmount and Age, these will appear as Targets.

Example Mapping Setup:

When executed, the workflow dynamically assigns runtime values to the Ruleset’s Facts, allowing the rules to be evaluated against real data.

Example Use-Case

1. Applying Knockout Rules for Borrowers

Scenario: A loan origination workflow applies knockout rules to identify applications that should be declined before entering the origination process. A Ruleset named "Borrower Knockout Rules" has been defined with the following Facts and Rules:

• Fact: LoanAmount → Rule: Must be greater than $100

• Fact: Age → Rule: Must be greater than 18

At runtime, loan application details are retrieved from a table using a Table Read Node and passed to the Ruleset Node.

Loan Application Data (Fetched at Runtime)

Ruleset Node Configuration

Workflow Execution & Ruleset Evaluation

During execution, the Ruleset Node evaluates the conditions based on runtime values:

• LoanAmount = 150 → ✅ Rule Passed (Approved)

• Age = 19 → ✅ Rule Passed (Approved)

Ruleset Node Output Format

🔹 What This Means: The borrower meets the knockout rule conditions, so the application should proceed to the next stage in the workflow.

Key Takeaways for Developers

✅ Automates Business Rule Enforcement – The Ruleset Node applies pre-defined logic dynamically to evaluate records against business rules in real-time.

✅ Seamless Workflow Integration – Developers can map runtime values from previous nodes (such as Table Read or API responses) directly into Ruleset Facts, ensuring rules are evaluated against live data.

✅ Enables Conditional Workflow Execution – By checking which rules pass or fail, developers can design workflows that automatically approve, reject, or trigger additional actions based on rule evaluation.

✅ Improves Decision-Making Efficiency – Instead of hardcoding business logic into workflows, rules are managed separately in a Ruleset, making updates easier and reducing workflow complexity.

✅ Flexible Data Mapping – Mappings allow data to be sourced from agent-level variables, previous node outputs, or static values, giving developers full control over runtime rule execution.

By leveraging the Ruleset Node, developers can implement complex business logic in a structured, reusable, and scalable manner, ensuring that workflows execute with rule-based intelligence. 🚀

What It Is?

Prompt engineering is the skill of crafting the right prompts to get the best responses from AI language models like chatbots or AI assistants. Since AI doesn't "think" like humans, the way you phrase your prompts significantly impacts the quality of the output.

Here's an analogy to help you understand prompt engineering: Imagine you're asking a librarian for help finding a book. If you simply say, "I want a book," the librarian might not know where to start. But if you say, "I'm looking for a historical novel set in the 19th century about a female protagonist," the librarian can provide a more specific and helpful response.

The same principle applies to prompt engineering. By providing clear, concise, and informative prompts, you can guide the AI model to generate more accurate, relevant, and creative responses.

How Does It Work?

Crafting effective prompts is crucial for getting the most out of AI language models. Let's delve deeper into the comparison between ineffective and effective prompts, and explore additional examples across various domains to illustrate the key principles of prompt engineering.

Ineffective vs. Effective Prompts: A Deeper Dive

The initial example showcases the stark contrast between a vague and a well-structured prompt. "Tell me about loans" is too broad and open-ended, yielding potentially overwhelming and unfocused results. In contrast, "What are the steps involved in obtaining a home loan? What documentation is required, and what criteria are used for approval?" demonstrates specificity, guiding the AI towards a targeted and informative response.

Examples

Historical Research:

• Ineffective: "Tell me about World War II."

• Effective: "Analyze the causes of World War II, focusing on the role of political ideologies and economic tensions."

Creative Writing:

• Ineffective: "Write a story."

• Effective: "Write a science fiction short story about a time traveler who accidentally alters the course of history."

Scientific Inquiry:

• Ineffective: "Explain climate change."

• Effective: "Discuss the impact of human activities on climate change, specifically the role of greenhouse gas emissions."

Technical Support:

• Ineffective: "My computer isn't working."

• Effective: "I'm encountering a blue screen error on my Windows 10 laptop. What troubleshooting steps can I take?"

The Power of Prompt Engineering

By mastering the art of prompt engineering, you can unlock the full potential of AI language models. Well-crafted prompts enable you to extract precise information, generate creative content, and explore complex topics with remarkable ease and efficiency. Remember, the quality of the AI's output is directly influenced by the quality of your input.

Here are some tips for effective prompt engineering:

• Clarity and Specificity: The foundation of a good prompt is clarity. Clearly articulate your request, leaving no room for ambiguity. Be specific about the format, style, and tone you expect in the response.

  Example:

  • Instead of: "Write about AI."

  • Use: "Write a 300-word article about the benefits of AI in healthcare, using a professional tone and including three examples of applications."

Advanced Prompting Techniques

• Prompt Chaining: Break down complex tasks into a sequence of simpler prompts, each building on the output of the previous one. Learn More

• Prompt Interpolation: Combine multiple prompts or prompt elements to generate more nuanced and sophisticated responses. Learn More

• Prompt Optimization: Use machine learning techniques to automatically optimize prompts for specific tasks or desired outcomes. Learn More

Ethical Considerations

• Bias Mitigation: Be mindful of potential biases in the AI model and take steps to mitigate them through careful prompt design and output evaluation.

• Harmful Content Prevention: Implement safeguards to prevent the AI from generating harmful or offensive content.

• Transparency and Accountability: Clearly communicate the limitations of the AI model and take responsibility for the outputs it generates.

From a developer's perspective, AI Agents and LLM-based apps like ChatGPT differ significantly in terms of architecture, capabilities, and use cases.

LLM-based apps are primarily focused on generating text based on a given prompt. They excel at tasks such as language translation, summarization, and content creation. However, their functionality is limited by their reliance on pre-trained models and their inability to interact with external systems or perform actions beyond generating text.

AI Agents, on the other hand, are designed to be more versatile and capable of performing a wider range of tasks. They can interact with their environment, make decisions, and take actions based on their goals. This is achieved through the integration of various components, such as perception modules, decision-making algorithms, and action execution mechanisms.

Why AI Agents Are the Next Step Beyond LLM-Based Apps

LLM-based apps have provided significant advancements in how users interact with software, but they have notable limitations. AI Agents address these limitations by offering context awareness, real-world action capabilities, and decision-making autonomy. Below is a detailed comparison:

1. Overcoming Limited Context with AI Agents LLM-Based Apps: Struggle with Context Retention

   • LLM-based apps typically rely on a stateless approach, meaning they process each user input independently.

   • While modern models support longer context windows, they still struggle with remembering past interactions over long sessions.

   How AI Agents Solve This

Key Differences Between AI Agents and LLM-Based Apps

Platform Shift & Evolution

• AI agents represent a major shift from traditional SaaS and LLM-based apps.

• Historically, software architecture evolved with platform changes (e.g., mainframes → cloud).

• Now, we’re moving from software-driven apps to AI-driven agents.

AI Agents vs. LLM-Based Apps

• LLM-based apps: These are applications that use large language models (LLMs) to enhance user interactions but still function as traditional apps.

• AI Agents: These are autonomous, goal-oriented systems that perform tasks on behalf of users with minimal human intervention.

Functionality Differences

• LLM-based apps require user input and respond accordingly.

• AI agents proactively take action based on intent, context, and automation.

The Future of Agents

• Agents will integrate deeply into workflows, replacing static SaaS interfaces.

• Instead of navigating multiple apps, users will interact with agents that dynamically execute tasks across various systems.

Implication for Developers

• Developers will need to build AI-native architectures instead of just embedding LLMs into traditional apps.

• AI agents will require new frameworks for decision-making, autonomy, and integration.

Key Takeaway for developers.

AI agents are not just chatbots or enhanced LLM-based apps—they are autonomous systems designed to replace traditional apps by executing actions dynamically. They consist of a reasoning engine for intent classification, inference, and task execution orchestration.

Overview

The Storage Read Node is the counterpart to the Storage Write Node, enabling developers to retrieve document metadata from storage using a storageId.

If developers need to fetch the actual document contents, they can use the documentId returned by this node in conjunction with the Fetch Document Node.

Common Workflow Pattern for Storage Read Usage

1️⃣ Retrieve the document metadata using the Storage Read Node (returns documentId). 2️⃣ Pass the documentId to the Fetch Document Node to get the actual file contents. 3️⃣ Process the document further using OCR, AI extraction, or other workflow steps.

🔹 Example Use-Case: In a loan processing system, an AI agent may need to retrieve previously stored financial statements. The Storage Read Node fetches the documentId, which is then passed to Fetch Document Node to extract and analyze the contents.

Configurations

Execution Flow:

1️⃣ The Storage Read Node receives a storageId as input. 2️⃣ It retrieves the corresponding documentId, which serves as the system’s primary key for document identification. 3️⃣ The documentId can then be used in a Fetch Document Node to retrieve the actual file.

Output Format:

• documentId → A unique identifier that represents the stored document.

Example Use-Cases

Use-Case 1: Retrieving KYC Documents for Loan Applications

A loan processing workflow needs to retrieve previously stored KYC documents (e.g., ID Proof, Address Proof) for verification before approval.

Configuration:

Execution Process:

1️⃣ The Storage Read Node retrieves the documentId associated with the stored KYC document. 2️⃣ The returned documentId is passed to the Fetch Document Node to retrieve the actual file. 3️⃣ The document is verified by AI models or manual review before final loan approval.

Use-Case 2: Fetching Financial Statements for Business Loan Assessment

A financial AI agent needs to retrieve and analyze past financial statements submitted during a business loan application.

Configuration:

Execution Process:

1️⃣ The Storage Read Node fetches the documentId for the financial statement. 2️⃣ The documentId is passed to the Fetch Document Node to retrieve the statement file. 3️⃣ The retrieved statement is processed by an AI model to extract financial insights.

Use-Case 3: Accessing Archived Invoices for Auditing

An enterprise finance system requires invoices stored in the system to be retrieved for tax filing and audits.

Configuration:

Execution Process:

1️⃣ The Storage Read Node retrieves the documentId for the stored invoice. 2️⃣ The documentId is passed to the Fetch Document Node, which fetches the actual invoice. 3️⃣ The invoice is reviewed by finance teams for tax compliance and auditing.

What is an Intent in Sub Agent?

An Intent represents a specific goal or action that an AI sub-agent in UPTIQ is designed to handle based on user input. It helps the reasoning engine understand and classify user queries to trigger the appropriate response or workflow.

Key Components of an Intent:

1. Intent Name

   • A unique identifier for the intent.

   • Should be easy to distinguish from other intents within the same sub-agent.

2. Intent Description

   • A clear and comprehensive explanation of the intent’s purpose.

   • Used by the UPTIQ reasoning engine to match user queries accurately.

   • Should be detailed yet simple for optimal LLM processing.

3. Intent Examples

   • Up to five sample queries that help the reasoning engine learn how users may phrase their requests.

   • These examples improve intent recognition accuracy.

How Intent Execution Works in UPTIQ AI Sub-Agents

When a user query is triggered, UPTIQ's reasoning engine follows a structured process to identify and execute the appropriate intent. Here's how it works:

1. User Query Processing

• The AI receives the user's input (e.g., "What is the total liability value?").

• The reasoning engine analyzes the query and compares it with defined intent examples to determine the best match.

2. Intent Matching

• If the query closely aligns with an existing intent's name, description, or examples, the system selects that intent.

• If no exact match is found, the engine attempts to generalize the query to the closest related intent.

3. Workflow Execution

• Once an intent is matched, the system automatically triggers the associated workflow.

• The workflow defines the next steps, such as:

  • Fetching relevant information from connected documents or databases.

  • Performing calculations or data extractions.

4. Response Generation

• The AI processes the workflow output and delivers a structured response to the user.

• The response is formulated based on the data retrieved and may include text, summaries, or extracted document details.

5. Continuous Learning & Improvement

• If users frequently ask queries that don’t match existing intents, developers can refine intents by:

  • Adding new examples to improve recognition.

  • Modifying intent descriptions for better clarity.

  • Creating new intents for uncovered scenarios.

Key takeaways for developers

✅ Purpose-Driven Design – Intents represent specific user goals or actions, enabling AI agents to deliver accurate and contextually relevant responses.

✅ Clear Naming for Easy Identification – Use unique, descriptive names to distinguish intents within sub-agents, improving clarity and organization in workflows.

✅ Comprehensive Descriptions – Provide clear, unambiguous descriptions to help the UPTIQ reasoning engine match user queries effectively.

✅ Use Examples for Precision – Add up to five examples of user queries to guide the reasoning engine and improve accuracy in intent recognition.

✅ Workflow Integration – Each intent is automatically linked to a workflow stub, allowing developers to define precise actions when an intent is matched.

✅ Iterative Refinement – Regularly update intent examples and descriptions to handle evolving user queries and improve performance over time.

✅ Modular Scalability – Intents can be expanded and refined without disrupting the functionality of other intents or workflows, ensuring scalability and flexibility.

By designing and maintaining well-structured intents, developers can create intelligent, responsive, and user-centric AI agents within UPTIQ.

How to create an intent in Sub Agent?

Overview

The Fetch Document Node is responsible for retrieving the actual content of a document using its documentId. While the Storage Read Node retrieves the documentId from storage, this node takes that documentId and returns a pre-signed URL, allowing developers to access the actual file securely.

🔹 Key Purpose: ✅ Fetch the actual document file that was previously stored. ✅ Generate a pre-signed URL for secure document access. ✅ Enable document viewing in a document viewer or AI processing pipeline.

Common Workflow Pattern for Fetch Document Usage

1️⃣ Retrieve the documentId using the Storage Read Node (or from previous workflow execution, variables) 2️⃣ Pass the documentId to the Fetch Document Node to obtain the pre-signed URL. 3️⃣ Use the pre-signed URL to display the document in a document viewer or process it further in an AI workflow.

🔹 Example Use-Case: A loan processing system retrieves a previously submitted financial statement. The Storage Read Node fetches the documentId, and the Fetch Document Node generates a pre-signed URL, which is used to display the document in a viewer for verification.

Configurations

Execution Flow:

1️⃣ The Fetch Document Node receives a documentId as input. 2️⃣ It generates a pre-signed URL, allowing secure access to the actual document. 3️⃣ The pre-signed URL can be used in a document viewer or passed to another workflow node for further processing.

Output Format:

• documentUrl → A secure, temporary link to access the document.

🔹 Important Note: ✔ The pre-signed URL is time-limited for security reasons. ✔ Developers should use the documentUrl immediately or refresh it when needed.

Example Use-Cases

Use-Case 1: Displaying a KYC Document in a Document Viewer

A loan processing workflow requires a loan officer to review a user’s submitted KYC documents (e.g., Passport, Address Proof) before approval.

Configuration:

Execution Process:

1️⃣ The Storage Read Node retrieves the documentId of the stored KYC document. 2️⃣ The Fetch Document Node generates a pre-signed URL for secure document access. 3️⃣ The document URL is displayed in a document viewer for manual review.

Use-Case 2: Extracting Information from a Stored Invoice

An AI-powered invoice processing system needs to extract key financial data from a previously uploaded invoice.

Configuration:

Execution Process:

1️⃣ The Storage Read Node fetches the documentId of the invoice. 2️⃣ The Fetch Document Node generates a pre-signed URL for secure access. 3️⃣ The document URL is passed to an OCR-based AI model, which extracts text and numeric values.

Use-Case 3: Automating the Retrieval of Financial Statements

A business loan underwriting system requires AI models to analyze past financial statements stored in the system.

Configuration:

Execution Process:

1️⃣ The Storage Read Node retrieves the documentId of the financial statement. 2️⃣ The Fetch Document Node generates a pre-signed URL for AI processing. 3️⃣ The document is analyzed by an AI model to assess financial health and eligibility.

Key Takeaways for Developers

✅ Retrieves Actual Document Content – Unlike the Storage Read Node, which returns only the documentId, the Fetch Document Node generates a pre-signed URL for secure document access.

✅ Enables Secure & Temporary Access – The generated pre-signed URL is time-limited, ensuring that documents remain protected and cannot be accessed indefinitely.

✅ Seamless Integration with AI & Document Processing – Works alongside AI models, OCR tools, and document viewers, allowing for automated processing, manual review, or AI-driven extraction.

By leveraging the Fetch Document Node, developers can efficiently retrieve, process, and display documents, making AI-driven workflows more powerful, secure, and intelligent. 🚀

Overview

The Workflow Node allows developers to attach an existing workflow as a subworkflow, enabling modularity, reusability, and simplified workflow management.

By using subworkflows, developers can: ✅ Avoid Duplication – Use the same workflow across multiple parent workflows. ✅ Simplify Complex Workflows – Break down large workflows into manageable steps. ✅ Improve Maintainability – Make debugging and updates easier by isolating logic into reusable subflows.

Configurations

How the Workflow Node Works

1️⃣ The parent workflow calls the selected subworkflow. 2️⃣ The input data is passed to the subworkflow for execution. 3️⃣ The subworkflow runs independently, processing its steps. 4️⃣ Once completed, the subworkflow returns the output to the parent workflow.

Example Use-Cases

Use-Case 1: Processing Invoices in a Finance Workflow

A finance automation workflow requires invoice processing as a separate, reusable step.

Configuration:

Execution Process:

1. The Process Invoice subworkflow is triggered with the input data.

2. It validates the invoice, checks compliance, and logs the transaction.

3. Once processed, it returns the final status to the parent workflow.

🔹 Why Use a Subworkflow? ✔ Keeps the parent workflow clean and focused. ✔ Allows multiple workflows to reuse the invoice processing logic. ✔ Easy to update invoice handling without modifying multiple workflows.

Use-Case 2: Loan Origination System

A loan application workflow needs to verify applicant details through a separate KYC (Know Your Customer) process.

Configuration:

Execution Process:

1. The Verify KYC subworkflow retrieves customer details.

2. It performs ID verification and credit checks.

3. The subworkflow returns an approval status to the parent workflow.

🔹 Why Use a Subworkflow? ✔ Standardizes KYC processing across multiple workflows. ✔ Allows quick updates to KYC rules without affecting multiple workflows. ✔ Improves workflow organization by isolating verification logic.

Use-Case 3: Data Enrichment Pipeline

A data processing workflow needs to clean and enrich incoming customer records before storage.

Configuration:

Execution Process:

1. The Data Cleansing subworkflow standardizes formats, removes duplicates, and enriches records.

2. It validates email addresses, phone numbers, and other key fields.

3. The subworkflow returns the cleaned record to the parent workflow for further processing.

🔹 Why Use a Subworkflow? ✔ Reduces repetitive data validation logic across workflows. ✔ Allows independent testing and debugging of the data cleansing process. ✔ Keeps the parent workflow lightweight and focused on orchestration.