Writing
Notes on the practical side of building with AI: retrieval, agents, streaming, databases and the plumbing that makes products reliable.
5 posts · 11 topics
Featured
Sep 8, 20263 min readAI / RAG / Postgres
Citations or it didn't happen: RAG that shows its sources
Showing 5 of 5 posts.
- Sep 8, 2026Citations or it didn't happen: RAG that shows its sourcesA retrieval pipeline is only as trustworthy as its receipts. Here is how I structure chunking, retrieval and prompting so every answer points back to the exact passage it came from.AIRAG
- Aug 21, 2026Streaming LLM responses in Next.js without the jankTokens should arrive smoothly, cancel cleanly and never freeze the page. A practical walkthrough of streaming from a route handler to a React client.Next.jsAI
- Jul 30, 2026MongoDB or Postgres: where should your vectors live?You probably don't need a separate vector database on day one. How I decide between Atlas Vector Search, pgvector and a dedicated store.DatabasesAI
- Jul 9, 2026Keep the agent loop boringTool-using agents fail in dull, predictable ways. Timeouts, budgets and typed tools matter more than clever prompts.AIAgents
- Jun 18, 2026The caching layers that make a storefront feel instantSub-second browsing comes from caching the right things in the right places: the CDN, the server and the browser, each with a clear job.PerformanceNext.js