DevNotes AI
A RAG-powered notes app that lets developers chat with their own technical notes and documentation.
An experiment in building a retrieval-augmented notes tool — write and organize technical notes as usual, then ask questions across all of them instead of searching manually.
Personal technical notes pile up fast and become hard to search meaningfully — keyword search misses context, and skimming old notes wastes time.
Notes are chunked and embedded on save, stored alongside their vectors, and a chat interface retrieves the most relevant chunks before asking an LLM to answer using only that context.
- Markdown note editor with automatic chunking on save
- Vector search over personal notes
- Chat interface answering strictly from retrieved context
- Source snippets shown alongside every answer
A Next.js app with API routes handling embedding generation and retrieval, Postgres with a vector extension for storage, and a simple prompt-construction layer that grounds every answer in retrieved snippets.
Getting chunking granularity right — chunks that were too large diluted retrieval relevance, and chunks that were too small lost surrounding context.
A hands-on understanding of how retrieval quality, not model choice, is usually the bottleneck in a RAG system.