AUTORAG

Self-Healing RAG KNOWLEDGE ENGINE AWS IMPACTXCHALLENGE

An improved Retrieval Augmented Generation (RAG) system with automatic self-healing capabilities that enhances answers by searching external sources when the base knowledge base has low confidence.

Features

• Self-Healing: Automatically searches Wikipedia and the web when trust scores are low
• Improved Text Cleaning: Enhanced noise removal and text processing
• Better Answer Quality: Smarter sentence extraction and deduplication
• FastAPI API: RESTful API for easy integration
• Intelligent Hybrid Retrieval: Combines base knowledge with external sources
• Caching: LRU cache for frequently queried topics
• Comprehensive Logging: Detailed logging for debugging and monitoring

Installation

1. Install dependencies:

pip install -r requirements.txt

Usage

Option 1: Run as FastAPI Server (Recommended)

python self_healing_rag.py

Or using uvicorn directly:

uvicorn self_healing_rag:app --reload --host 0.0.0.0 --port 8000

Then visit:

• API Documentation: http://localhost:8000/docs (Interactive Swagger UI)
• Alternative Docs: http://localhost:8000/redoc
• Root: http://localhost:8000/

Option 2: Run Demo Script

python run_rag.py

This will initialize the system and run some example queries.

API Endpoints

POST /query

Query the RAG system.

Request Body:

{
  "query": "What is quantum computing?",
  "threshold": 0.5,
  "max_results": 5,
  "use_healing": true
}

Response:

{
  "query": "What is quantum computing?",
  "answer": "...",
  "before_answer": "...",
  "trust_score_before": 0.26,
  "trust_score_after": 0.81,
  "healing_triggered": true,
  "healing_successful": true,
  "sources_used": ["Base Knowledge Base", "Wikipedia: Quantum_computing"],
  "timestamp": "2025-01-XX..."
}

POST /query/demo

Same as /query but returns formatted output for easy viewing.

GET /health

Health check endpoint.

GET /

API information.

Improvements Made

1. Better Text Cleaning:

   • Enhanced noise pattern removal
   • Better HTML/JavaScript filtering
   • Improved sentence extraction
   • Deduplication of similar sentences

2. Improved Self-Healing Logic:

   • Prioritizes Wikipedia API for cleaner results
   • Better web scraping with content area detection
   • Smarter hybrid retrieval (weighted combination)
   • Improved trust score thresholds

3. Better Answer Quality:

   • Removes markdown and HTML artifacts
   • Better sentence boundary detection
   • Minimum length filtering for quality
   • Duplicate removal

4. Performance & Reliability:

   • LRU caching for queries
   • Better error handling
   • Comprehensive logging
   • Timeout handling for web requests

5. API Features:

   • FastAPI with automatic documentation
   • Request validation with Pydantic
   • Detailed response models
   • Demo endpoint with formatted output

Example Queries

Try these queries to see the self-healing in action:

• "What is quantum computing?"
• "What are the latest AI regulations in Europe?"
• "How does OAuth authentication work?"
• "Explain machine learning algorithms"
• "What is blockchain technology?"

Configuration

You can adjust the following parameters:

• threshold: Trust score below which healing is triggered (default: 0.5)
• max_results: Maximum number of results to return (default: 5)
• use_healing: Enable/disable self-healing (default: true)

Requirements

See requirements.txt for all dependencies. Main dependencies:

• FastAPI
• sentence-transformers
• faiss-cpu
• datasets
• duckduckgo-search
• beautifulsoup4
• requests

Notes

• First run will download the sentence transformer model (~80MB)
• First run will download the Wikitext dataset (~500MB)
• Initialization takes a few minutes to create embeddings
• Web scraping may take a few seconds per query