Product · AI
Docly
AI document chat: upload several documents, process them in the background, and get answers grounded in their content.
- My role
- Solo — architecture, backend, frontend
- Stack
- Node.jsExpressMongoDBAWS S3BullMQRedisOpenAI Embeddings
The problem
Chatting with documents breaks down when uploads are processed inline: large files block requests and answers drift away from the source. Docly separates ingestion from chat so both stay reliable.
What I built
- Multi-document upload to AWS S3, with processing queued through BullMQ on Redis so the API never blocks on large files.
- An ingestion pipeline that chunks documents and generates 1536-dimensional OpenAI embeddings for vector retrieval.
- Semantic retrieval feeding an LLM response layer, so answers are grounded in the uploaded documents.
- Conversational context built from the latest 5 messages to keep follow-up questions coherent.