# WTechNI

> Practical AI engineering: protocol design, retrieval and evaluation.


## Posts

- [AGENTS.md: Give Coding Agents Real Repo Context](https://dev.wtechni.com/agents-md-guide/): What to put in AGENTS.md, what to keep out, and how to measure whether repo-level context actually changes how your coding agent behaves.
- [Build an LLM Eval Harness That Runs in CI](https://dev.wtechni.com/llm-eval-harness-ci/): Forty cases, a scorer, a baseline and a GitHub Action. How to catch AI regressions that unit tests miss, without paying for an eval platform.
- [RAG Chunking: Choose Chunk Size With Evals](https://dev.wtechni.com/rag-chunking-strategy-evaluation/): There is no best chunk size. Build a 30-question eval set, index at three sizes, and let recall@k pick the chunker instead of a blog post.
- [Build an MCP Server in Python with FastMCP](https://dev.wtechni.com/build-mcp-server-python-fastmcp/): A working MCP server in twenty lines, then the parts that only break in production: tool design, validation, error messages, testing and security.
- [MCP 2026-07-28: What Changed and How to Migrate](https://dev.wtechni.com/mcp-2026-07-28-stateless-spec-migration/): MCP's 2026-07-28 revision drops sessions and the initialize handshake. What breaks, what replaces it, and the exact migration order for your server.

## Pages

- [Home](https://dev.wtechni.com/)
- [Articles](https://dev.wtechni.com/articles/)


[comment]: # (Generated by Hostinger Tools Plugin)