Writing
4 posts on AI, security, and web performance.
- How a question finds its answer The read side of the pipeline: turn a question into a vector, find the nearest chunks with one SQL query, optionally rerank them, wrap them in a prompt that forces the model to stay honest, and stream back an answer that cites its sources.
- How a document becomes searchable How a markdown scheme document becomes searchable rows in pgvector: paragraph-aware chunking that preserves citations, a provider-agnostic embedding layer, and the fixed vector column dimension that will bite you if you ignore it.
- What is a vector database? The one idea the whole series rests on, from scratch: how text turns into coordinates, why the nearest point is the most similar in meaning, and what the index is really doing. No prior vector knowledge assumed.
- What we're building, and why A build-along series on a retrieval-augmented Q&A system over Indian government schemes: what we're building, why grounding and citations matter more than the model, and the pipeline the rest of the series puts together piece by piece.