backend

Chat follow-up questions without server sessions: history lives in localStorage, gets resent per ask, and the backend rebuilds the prompt under a char budget.

The blog's AI chat now answers 'what are you working on' from local commit history: a narrow port, a version namespace bump, and term mining.

When the LLM router died, a meta fallback event arrived and the visitor still got an answer. Here's the three-part chain behind it.

Typing hai used to kill the AI search with no_results. The culprit was MariaDB's fulltext word-length floor, not a broken prompt.

One tiny field on the list page nearly cost a query per article. The fix is one batch IN query: two queries per page, not 1 + N.

An optional locale filter froze live search on the default locale. The culprit: bindparams validation that fires at construction, not execution.

Replace LIKE with a MariaDB FULLTEXT index, then add LLM query expansion: two layers that keep working when the index or the LLM fails.

The semantic related layer for 80 articles turned out to be a Python dict and a cosine loop, no numpy and no vector database.

Snippet changes used to mean an edit, a commit, and a deploy wait. A whitelisted settings endpoint turns that into one PUT request.