IntelliSearch

Semantic search with hybrid ranking, RAG answers, and fast file ingestion.

Live Demo → https://intellisearch-eight.vercel.app

Highlights

• ⚡ Processing 4,000+ documents
• 🔍 3 search modes (Semantic, Hybrid, RAG)
• 🤖 AI-powered answers with GPT-4o-mini
• ⏱️ Sub-500ms retrieval speed
• 💾 Redis caching for cost optimization

Architecture

User → React Frontend (Vercel)
         ↓
    FastAPI Backend (Railway)
         ↓
    ┌────┴────┬─────────┐
    ↓         ↓         ↓
OpenAI    Pinecone   Redis
(Embed)   (Vectors)  (Cache)

Tech Stack

Frontend

• React
• TypeScript
• Tailwind CSS
• Vite

Backend

• Python
• FastAPI
• Uvicorn

AI/ML

• OpenAI

Database + Cache

• Pinecone
• Redis

Deploy

• Railway
• Vercel

Features

• Semantic search with OpenAI embeddings (text-embedding-3-small)
• Hybrid search (vector similarity + keyword overlap)
• RAG search with gpt-4o-mini for grounded answers
• File uploads for PDF, DOCX, TXT, and Markdown
• Chunking for large documents with parent/child metadata
• Pinecone vector store with auto-index creation
• Redis caching for embeddings and search responses (optional)
• Batch ingestion and bulk upload script

Quick Start

1. Backend

cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

2. Frontend

cd frontend
npm install
npm run dev

3. Environment

Create backend/.env:

OPENAI_API_KEY=your_openai_key
PINECONE_API_KEY=your_pinecone_key
REDIS_URL=redis://localhost:6379

Optional frontend override:

VITE_API_BASE_URL=http://localhost:8000

API Documentation

Health

• GET /health

Ingestion

• POST /documents/ingest
  • Body: { "content": "text", "metadata": {} }
  • Query: chunk=true
• POST /documents/batch-ingest
  • Body: [{"content": "...", "metadata": {...}}, ...]
  • Query: chunk=true

File Uploads

• POST /documents/upload
  • Form file: file
  • Query: chunk=true
  • Query: metadata (JSON string)
• POST /documents/upload-batch
  • Form files: files[]
  • Query: chunk=true
  • Query: metadata (JSON string)

Search

• POST /search
  • Body: { "query": "text", "top_k": 5 }
• POST /hybrid-search
  • Body: { "query": "text", "top_k": 5, "vector_weight": 0.7 }
• POST /rag-search
  • Body: { "query": "text", "top_k": 3 }

Document Management

• GET /documents
  • Query: limit (default 20), offset (default 0)
• GET /documents/{id}
• DELETE /documents/{id} (also deletes chunk children)

Bulk Upload Script

The script backend/bulk_upload.py generates synthetic content and uploads in batches.

cd backend
python bulk_upload.py

It targets the Railway production URL by default. Update API_URL in the script for local ingestion.

Deployment (Railway)

Docker-based deployment via Dockerfile + railway.json. The container runs:

uvicorn app.main:app --host 0.0.0.0 --port ${PORT:-8000}

Required Railway environment variables:l

• OPENAI_API_KEY
• PINECONE_API_KEY
• REDIS_URL (optional)

Author

• Name: Gurraj Singh