DocsGPT is a retrieval-augmented assistant that answers documentation questions, summarizes technical material, and returns contextual code snippets, built on LangChain with ChromaDB as the vector store.
How it works
Documentation is chunked and embedded into ChromaDB. At query time, relevant chunks are retrieved and passed to an LLM alongside the user's question, grounding the response in the actual documentation rather than the model's general knowledge — which matters most for fast-moving libraries where the model's training data is stale relative to the current API.
Related publication
The design and evaluation of this system is written up in the accompanying publication, currently under peer review.