README.md

Triluxo Tech Assignment

This project implements a custom chatbot using Langchain, Google Gemini AI, and Flask. It extracts course data from Brainlox, creates embeddings for semantic search, and exposes a RESTful API for conversational interactions.

Overview

The chatbot works by:

1. Fetching Data: It loads course data from a specified Brainlox URL using Langchain's WebBaseLoader.
2. Parsing Content: It parses the HTML and extracts course titles and descriptions.
3. Splitting Text: Long texts are split into smaller chunks for better processing.
4. Creating Embeddings: It converts the text chunks into numerical embeddings and stores them in a Chroma vector database.
5. Serving the API: A Flask RESTful API is set up to handle user queries, retrieve context, and generate responses using Google Gemini.
6. Managing Sessions: It manages user sessions to provide context-aware responses.
7. Rate Limiting and Caching: Applies rate limiting to the API endpoint to prevent abuse and implements caching to improve response times.

Example I/O

Input API Request

```bash
curl -X POST \
  -H "Content-Type: application/json" \
  -d '{"message": "What courses are available?"}' \
  http://0.0.0.0:5000/chat
```

Output

Prerequisites

Before you start, make sure you have the following:

• Python 3.8+

• A Google Gemini API Key: Get one from the Google AI Studio.

• Environment Variables Set Up: Create a .env file in the project root, and add your API key and other configurations. Example .env (replace placeholders with your actual values):

  GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
  DATA_URL="https://brainlox.com/courses/category/technical"
  CHUNK_SIZE=1000
  CHUNK_OVERLAP=200
  EMBEDDINGS_MODEL="sentence-transformers/all-MiniLM-L6-v2"
  TEMPERATURE=1
  TOP_P=0.95
  TOP_K=40
  MAX_OUTPUT_TOKENS=8192
  RESPONSE_MIME_TYPE="text/plain"
  MODEL_NAME="gemini-2.0-flash-exp"
  SIMILARITY_SEARCH_K=3
  FLASK_DEBUG=True

  • GEMINI_API_KEY: REQUIRED: Your Google Gemini API key.
  • DATA_URL: URL of the source data (default: https://brainlox.com/courses/category/technical).
  • CHUNK_SIZE: Size of text chunks (default: 1000).
  • CHUNK_OVERLAP: Overlap between chunks (default: 200).
  • EMBEDDINGS_MODEL: Hugging Face embeddings model (default: sentence-transformers/all-MiniLM-L6-v2).
  • TEMPERATURE: Sampling temperature for the model (default: 1).
  • TOP_P: Top-p sampling parameter (default: 0.95).
  • TOP_K: Top-k sampling parameter (default: 40).
  • MAX_OUTPUT_TOKENS: Maximum tokens in a response (default: 8192).
  • RESPONSE_MIME_TYPE: Response type, used to return a plain text response (default: text/plain).
  • MODEL_NAME: Name of the Gemini model (default: gemini-2.0-flash-exp).
  • SIMILARITY_SEARCH_K: The number of documents to retrieve from vector DB (default: 3).
  • FLASK_DEBUG: Enable Flask debug mode (default: True).

Setup and Installation

1. Clone the repository:

   git clone https://github.com/Kaos599/Triluxo-Technologies-Assignment.git

2. Install dependencies:

   pip install -r requirements.txt

3. Set up .env file Create a .env file in the project root, and add all environment variables specified above, make sure to replace all the placeholders with your correct values.

Running the Application

1. Run the Flask app:

   python app.py

   The application will start on http://0.0.0.0:5000.

API Usage

The application provides the following API endpoints:

1. /chat (POST)

Send messages to the chatbot and receive responses.

• Method: POST

• Headers:

  • Content-Type: application/json

• Request Body:

  {
      "message": "Your query here",
      "session_id": "Optional session id"
  }

  • message: (required, string) The user's message.
  • session_id: (optional, string) Session ID for ongoing conversations.

• Response Body:

  {
    "response": "Chatbot's response message",
    "session_id": "The active session ID",
    "sources": ["list of urls of the data sources used to generate the response"]
  }

  • response: (string) The chatbot's response.
  • session_id: (string) Session ID of the current session.
  • sources:(list of strings) A list of URL source where the data was found to generate the answer.

• Example curl Request:

  curl -X POST \
    -H "Content-Type: application/json" \
    -d '{"message": "What courses are available?"}' \
    http://0.0.0.0:5000/chat

2. /health (GET)

Check the status of the application.

• Method: GET

• Response Body:

  {
      "status": "ok",
      "model": "gemini-2.0-flash-exp"
  }

  • status: (string) The status, which should be "ok".
  • model: (string) The name of the model in use.

• Example curl Request:

   curl http://0.0.0.0:5000/health

Important Notes

• The vector database is saved locally in the chroma_db directory. It will only be recreated if the REFRESH_VECTOR_STORE environment variable is set to "true" (not in the example env provided).
• The data loading is enhanced to parse HTML and extract course information. The parsing logic in the load_data function might need updates if the target website changes.
• Rate Limiting: The /chat endpoint is rate-limited to 10 requests per minute.
• Error Handling: Detailed logs are provided.
• Session Management: Sessions are stored in memory, not persistently.
• Caching: Responses are cached for 5 minutes to improve response times.