Chat Assistant with LLM & Rasa

Duration: Dec 2022 - Feb 2023

Overview

Built a comprehensive 4-layer chatbot system integrating Mistral 7B, Rasa Open Source, and FastAPI. The system features real-time communication, automated training pipelines, and a modern Next.js frontend.

Key Features

• Architected a 4-layer chatbot system using Mistral 7B, Rasa Open Source, and FastAPI
• Created an LLM server using LangChain to integrate and utilize Mistral 7B efficiently
• Developed custom connectors for REST communication with streaming handler
• Implemented automated pipeline for training data updates
• Designed intuitive chatbot UI using Next.js

Technical Stack

{
  "backend": {
    "llm": "Mistral 7B",
    "framework": "Rasa Open Source",
    "api": "FastAPI",
    "integration": "LangChain"
  },
  "frontend": {
    "framework": "Next.js",
    "communication": "REST API"
  }
}

Architecture

The system comprises four distinct layers:

1. User Interface (Next.js)
2. API Gateway (FastAPI)
3. LLM Processing (Mistral 7B)
4. Conversational AI (Rasa)

Implementation Details

LLM Integration

• Leveraged LangChain for efficient model utilization
• Implemented streaming for real-time responses
• Optimized resource usage for production deployment

Custom Connectors

• Built REST-based communication protocol
• Developed streaming handler for real-time updates
• Ensured seamless component integration

Training Pipeline

• Automated data processing workflow
• Continuous model improvement system
• Enhanced accuracy through iterative updates