Repository files navigation

This system uses a GitHub-hosted LLM (via the GitHub model API) to classify and respond to 12+ common customer support inquiries, integrates with Intercom for automated replies, and falls back to macros for reliability.

• Real AI-powered intent classification using a GitHub-hosted LLM (seeconfigs/config.py for model and endpoint).
• Handles 12+ common support intents (seeai_config/pylon_model_config.json).
• Macro-based fallback for reliability.
• Intercom integration for automated response delivery.
• Easily extensible and testable.

1. Install dependencies:

   pip install -r requirements.txt

2. Environment variables:

   • GITHUB_TOKEN: Your GitHub access token (with model access enabled)
   • INTERCOM_ACCESS_TOKEN: Your Intercom personal access token
   • INTERCOM_ADMIN_ID: Your Intercom admin ID

   You can set these in your shell or in a.env file. The project usespython-dotenv to load environment variables automatically.

   Example.env file:

   GITHUB_TOKEN=your_github_access_tokenINTERCOM_ACCESS_TOKEN=your_intercom_tokenINTERCOM_ADMIN_ID=your_admin_id

3. Configuration files:

   • ai_config/pylon_model_config.json: List of supported intents and model settings
   • fallback_macros/intercom_macros.json: Macro responses for each intent
   • data/sample_queries.json: Example queries for training/evaluation
   • configs/config.py: Model, endpoint, and generation parameters

• Run the system interactively (for local usage):

  python main.py

  This will prompt you for a customer query and a conversation ID, classify the intent, generate a response, and (if configured) send it to Intercom.

• Use the PylonAI class in your code:

  fromai_config.pylon_aiimportPylonAIai=PylonAI("ai_config/pylon_model_config.json")ai.handle_query("I want a refund for my last purchase.",conversation_id)

  The system will classify the query, generate a response, and send it via Intercom.

• The code usesfrom intercom.client import Client and theconversations.reply ormessages.create method to send replies to Intercom conversations.
• If you are running locally and do not want to send real messages, use dummy values for the environment variables.
• If the environment variables are not set, the system will use empty strings, which will not connect to Intercom.

• Run all tests with:

  pytest tests

• Thepytest.ini file ensures the project root is in the Python path for imports.
• Seetests/ for test cases, including performance, privacy, and security.
• Requirements and QA deliverables:docs/QA_Assignment_Deliverables.md

• ModuleNotFoundError:
  • Ensure you are running from the project root and that dependencies are installed.
  • Thepytest.ini file should resolve most import issues for tests.
• Intercom import errors:
  • The correct import isfrom intercom.client import Client for this project version.
  • Ensure you haveintercom==0.2.3 installed (as specified inrequirements.txt).
• Environment variable issues:
  • Make sure your.env file is present and correct, or export variables in your shell.
  • The system will use empty strings if variables are missing, which disables real Intercom integration.

• If the GitHub model API is unavailable, the system falls back to keyword matching and macro responses.
• Ensure all environment variables are set before running for full functionality.
• For local testing, you can use dummy values for all environment variables.

• Add new intents topylon_model_config.json andintercom_macros.json.
• Update training data insample_queries.json for better accuracy.

1. Log in to GitHub and ensure you have access to the GitHub model API (see your organization or repository settings for model access).
2. Go to Settings > Developer settings > Personal access tokens.
3. Generate a new token with the required scopes for model access.
4. Use this asGITHUB_TOKEN in your.env file.

1. Log in to Intercom and go to the Developer Hub.
2. Create a new app (Internal integration).
3. In your app, go to Authentication and generate a Personal Access Token.
4. Use this asINTERCOM_ACCESS_TOKEN in your.env file.

For more, see the QA deliverables and code comments.