>_blog · 4
Articles
I write about things I have built and shipped myself: how they work, where the limits are, and when they are worth using.
- MCP: how to give AI access to your products and data What the Model Context Protocol is, how it works after the 2026-07-28 revision, how to write your own MCP server, protect it with authorization and keep the model from doing too much.
- RAG from scratch: how to make an LLM answer from your documents What RAG is, when you need it, how embeddings, chunking, hybrid search and reranking work, and how to build a first working version on MongoDB step by step.
- Evals: how to check that your AI works beyond the demo What evals are, what they consist of, how to grade answers with code, an LLM judge and a human, how to build your own dataset and which tools exist for this in 2026.
- How to share a password without leaving it in chat history forever Why I built sec.teqh.org: the secret is encrypted right in the browser, the server stores only ciphertext, and the record deletes itself after a set time or number of views.