Built a RAG application that ingests user-uploaded documents and websites to enable context-aware question answering over multi-modal data (text, images, tables).
Architected the full knowledge pipeline from ingestion to answer generation, covering parsing, chunking, embed- ding, retrieval, reranking, and citation-backed responses.
Developed agentic workflows and asynchronous job processing to keep long-running ingestion tasks off the server and the product responsive under load.
Integrated secure user authentication and provisioning, and wired the system to handle large-scale unstructured content reliably from upload to storage.
Strengthened observability and debugging with end-to-end trace analysis, then shipped the application as a modular containerized stack on AWS with a PostgreSQL backend.
Tech Stack: Next.js, TypeScript, TailwindCSS, Clerk — Python, FastAPI, PostgreSQL, Amazon S3, Redis, Celery — LangSmith, LangChain, LangGraph, Unstructured.io — Docker, AWS
Built with