Open-source fine-tuning tools have matured. Fine-tuning a small model locally with consumer GPU is now feasible. This is practical experience: which tools to use, data preparation, common pitfalls, and when fine-tuning is worth it.

RAG system evaluation is hard. RAGAS, Trulens, LLM-as-Judge—this article introduces practical RAG evaluation methods.

After LLM service goes live, monitoring Prompt/Response patterns is key to finding issues. This article covers LLM observability implementation.

GitFlow is dead, Trunk-Based Development is the right approach for modern dev teams. This article explains migration strategy.

Redis 7 introduced Redis Stack (search, JSON, graph DB). This article explains Redis 7 key changes and selection advice.

MCP (Model Context Protocol) has been out for a year. This is my experience using MCP to build real workflows: pitfalls encountered, most valuable usage patterns, and scenarios where MCP is反而负担.