Back to all selected work
Document Intelligence · RAG·2024·live
Quill
Upload a PDF and ask questions grounded in its contents using embeddings, vector retrieval, and streamed AI answers.
Next.jsTypeScriptLangChainPineconeGeminiPrismaStripe
Overview & Context
Quill enables users to chat with complex PDF documents in real time, leveraging Pinecone vector search and Gemini streaming.
The Problem
Reading through 50+ page legal, financial, or academic documents to extract specific clauses or figures is tedious and prone to human oversight.
The Architectural Approach
Implemented UploadThing PDF uploads, LangChain page chunking, Gemini vector embeddings stored in Pinecone, and streamed conversational retrieval. Includes Kinde authentication, Prisma/PostgreSQL state management, and Stripe plan tiers with file size and upload limits.
Key Technical Highlights
- Streamed AI responses with Pinecone similarity retrieval over PDF chunks
- UploadThing file ingestion with LangChain document splitting
- Kinde auth, Prisma database, and tiered Stripe subscriptions
Next Case Study
View projectImagine
A prompt-to-image web app with account credit balances and Razorpay credit purchases.