A high-performance Retrieval-Augmented Generation (RAG) platform designed for intelligent PDF querying, document analysis, and conversational AI grounded in custom document corpora.
Key Highlights & Features
- Local Vector Retrieval & Embeddings: Generates 384-dimensional local embeddings client-side and stores vector indexes in browser IndexedDB for lightning-fast retrieval without constant server roundtrips.
- Streaming Responses & Low Latency: Integrated Groq API for sub-250ms Time-To-First-Token (TTFT) and streaming model responses with precise page-level citations.
- Hybrid Retrieval: Combines local context retrieval with live web search when documents lack sufficient domain information.
- Container Optimization: Built resumable PDF ingestion and a multi-stage Docker production pipeline, slashing container image size from 1.91 GB to 475 MB (a 75% reduction).
- Edge Infrastructure: Integrated Cloudflare Workers for edge API rate limiting, caching, and resilient request routing.
Tech Stack
- Frontend / Framework: Next.js, React, Tailwind CSS, TypeScript
- Database & Storage: Supabase, PostgreSQL with
pgvector - AI & Embeddings: Groq Cloud, Ollama, Transformers.js (384-dim embeddings)
- Deployment: Docker Multi-Stage, Cloudflare Workers