Skip to main content

Building Custom LLM Applications

Unlike the Quick Start and Templates options, the Create LLM App interface gives you complete control over every aspect of your AI application. This builder allows you to construct sophisticated applications by connecting modules, configuring specific parameters, and optimizing performance to match your exact requirements.

Application Builder Interface​

The Create LLM App interface consists of these key areas:

  1. Module Selector (left sidebar): Choose which modules to add to your application
  2. Canvas (center): Visual workspace where you arrange and connect modules
  3. Module Configuration (right): Settings for each selected module
  4. Application Controls (top): Tools for application naming, saving, and creation

Connecting Modules​

The foundation of every application is the connection between different functional modules. These connections determine how data flows through your application, from initial input to final output.

Understanding the Module Flow​

Every Queryloop application follows a logical processing flow:

  1. Input Processing: Receive and interpret user requests
  2. Information Retrieval (optional): Find relevant data from your documents
  3. Response Generation: Create AI-generated responses
  4. Output Delivery: Return the formatted result to the user

Starting with Input Nodes​

All applications begin with the Input Node, which serves as the entry point for user queries and application configuration:

  • Problem Statement: Defines the purpose and scope of your application
  • Golden Responses: Provides example question-answer pairs
  • Advanced Options: Configures tone, citations, and other response characteristics

Problem Statement Configuration​

The problem statement helps the LLM understand what your application is designed to do:

  1. Click on the Problem Statement field
  2. Enter a clear description of your application's purpose
  3. Click "Advanced Options" to configure additional settings
  4. Click "Generate Prompt" after making any changes to update the system prompts

Important: Whenever you modify the problem statement, always click "Generate Prompt" afterward to ensure your changes are properly incorporated into the application's behavior.

Golden Responses Configuration​

Golden responses establish expected question-answer patterns:

  1. Select your preferred QnA option:
    • "I have golden QnA": Add your own example pairs
    • "I have sample questions only": Provide questions without answers
    • "I don't have any golden QnAs": Proceed without examples
  2. Click "Add Golden Responses" to enter your examples
  3. Enter question-answer pairs that represent ideal interactions

Adding Retrieval Modules​

Retrieval modules enable your application to search through documents for relevant information:

  1. Click the blue connection dot on the Input Node
  2. Select "Retrieval" from the module options
  3. Configure retrieval settings:
    • Upload Dataset: Add documents for the application to access
    • Choose Dataset Type: Select structured, unstructured, or unstructured with metadata
    • Set Confidentiality: Determine who can access the documents
    • Configure additional retrieval parameters (chunk size, metric type, etc.)

Dataset Configuration​

When uploading datasets, ensure you select the appropriate type and format:

  • Text Data: For unstructured text documents (.pdf, .docx, .txt, .md)
  • Structured Data: For tabular or organized data (.csv, .xlsx)
  • Confidentiality Levels: Public, Private, or Confidential

Adding Generation Modules​

Generation modules produce AI responses using either retrieved information or direct user queries:

  1. Click a blue connection dot (either from Input or Retrieval)
  2. Select "Generation" from the module options
  3. Configure generation settings:
    • Select LLM: Choose which language model to use
    • Configure Temperature: Set creativity level
    • Set Token Limits: Control response length
    • Select Prompt Type: Zero-shot, few-shot, chain of thought, etc.

Naming Your Application​

Giving your application a meaningful name helps with organization and identification:

  1. Click on the application name field at the top of the interface (default is your organization name + application number)
  2. Enter a unique, descriptive name
  3. Press Enter to save the name

Note: Application names must be unique within your organization. Descriptive names make it easier to identify applications in the Experiments and Deployment tabs.

Dataset Formats and Requirements​

Different functionality requires specific data formats. For optimal performance, follow these guidelines:

Golden Responses Dataset​

  • File Format: CSV
  • Required Headers:
    • question: The query text
    • answer: The expected response

Fine-tuning Datasets​

  • File Format: CSV
  • Required Columns:
    • system_prompt: Instructions for the model
    • query: The input text
    • response: The desired output

Prompt Few Shot Dataset​

  • File Format: CSV
  • Required Columns:
    • question: Example query
    • answer: Example response

Prompt Dynamic Few Shot Dataset​

  • File Format: CSV
  • Required Columns:
    • question: Example query
    • answer: Example response

Validating and Creating Your Application​

Before finalizing your application, verify all connections and configurations:

  1. Use the "Check" option in the left sidebar to validate your application
  2. Review any warnings or errors
  3. Make necessary adjustments
  4. Click the "Create" button in the top-right corner

Upon successful creation, you'll be directed to the Experiments tab, where you can optimize and test your application.

Best Practices for Application Creation​

To ensure the best results when building your application:

  1. Start with a Clear Problem Statement: Define precisely what your application should do
  2. Add Comprehensive Golden Responses: Include diverse examples covering various use cases
  3. Test Before Scaling: Build a minimal version first, then expand functionality
  4. Regenerate Prompts After Changes: Always update generated prompts when modifying parameters
  5. Use Appropriate Dataset Types: Match your data format to the right module configuration

Troubleshooting Common Issues​

Problem Statement Changes Not Reflected​

  • Solution: Click "Generate Prompt" after making changes to update the system instructions

Modules Not Connecting​

  • Solution: Ensure you're clicking on the blue connection dots, not elsewhere on the module

Dataset Upload Failures​

  • Solution: Verify your file format matches the required structure for the selected function

Missing Golden Responses​

  • Solution: Add responses directly through the Input Node's Golden Responses section

Next Steps​

After creating your application:

  1. Proceed to the Experiments tab to optimize performance
  2. Test different parameter combinations
  3. Select the best-performing configuration
  4. Deploy your application for production use

By utilizing the full capabilities of the Create LLM App interface, you can build sophisticated AI applications tailored to your specific requirements with unparalleled flexibility and control.