avinn555/typescript-rag

TechDocs TypeScript Book RAG API

A Retrieval Augmented Generation (RAG) system that provides precise answers to questions about TypeScript by searching through the official TypeScript Book.

🎯 Features

• —Semantic Search: Uses embeddings to find relevant content
• —Context-Aware Answers: Generates precise answers using retrieved documentation
• —Source Attribution: Returns source references for transparency
• —Fast Retrieval: Pre-computed embeddings for quick searches

🚀 API Usage

Base URL

https://YOUR_USERNAME-typescript-rag.hf.space

Endpoints

GET /search?q=question

Main search endpoint for querying the TypeScript Book.

Parameters:

• —q (required): Your question about TypeScript

Response:

{
  "answer": "string containing the answer",
  "sources": "referenced documentation files"
}

Example Requests:

1. 1.Fat Arrow Syntax

curl "https://YOUR_USERNAME-typescript-rag.hf.space/search?q=What%20does%20the%20author%20affectionately%20call%20the%20=%3E%20syntax?"

Expected answer includes: fat arrow

1. 1.Boolean Conversion Operator

curl "https://YOUR_USERNAME-typescript-rag.hf.space/search?q=Which%20operator%20converts%20any%20value%20into%20an%20explicit%20boolean?"

Expected answer includes: !!

GET /

Returns API information and status.

GET /health

Health check endpoint showing:

• —API configuration status
• —Number of chunks loaded
• —Embedding generation status

🔑 Configuration

Required Secret:

• —AIPIPE_API_KEY: Your AI Pipe API token

Setup:

1. 1.Go to Settings → Repository secrets
2. 2.Add secret: AIPIPE_API_KEY
3. 3.Space will automatically rebuild

🛠️ How It Works

1. 1.Content Ingestion: Fetches TypeScript Book chapters from GitHub
2. 2.Chunking: Splits content into semantic paragraphs
3. 3.Embedding Generation: Creates vector embeddings using Jina AI
4. 4.Retrieval: Finds top-K most relevant chunks for each query
5. 5.Answer Generation: Uses GPT-4o-mini to synthesize precise answers

📚 Data Source

Content from: TypeScript Deep Dive Book by Basarat Ali Syed

🧪 Testing

import requests

base_url = "https://YOUR_USERNAME-typescript-rag.hf.space"

# Test query
response = requests.get(f"{base_url}/search", params={
    "q": "What does the author affectionately call the => syntax?"
})

print(response.json())

📊 Technical Stack

• —Framework: FastAPI
• —Embeddings: Jina AI (jina-embeddings-v3)
• —LLM: GPT-4o-mini via OpenRouter
• —Retrieval: Cosine similarity search
• —Source: TypeScript Book (GitHub)

⚡ Performance

• —Startup: ~30-60 seconds (fetching & embedding content)
• —Query Response: ~2-5 seconds
• —Chunks Loaded: ~200-500 documentation chunks
• —Top-K Retrieval: 5 most relevant chunks per query

🔒 Security

• —API key stored as repository secret
• —CORS enabled for accessibility
• —No sensitive data stored or logged

📝 Example Questions

• —"What does the author affectionately call the => syntax?"
• —"Which operator converts any value into an explicit boolean?"
• —"What is the difference between let and const?"
• —"How do arrow functions handle the this keyword?"
• —"What are ambient declarations in TypeScript?"

📄 License

Educational/demonstration project for TechDocs Inc.