Ask Rubén

How it works

Every page on this site is a markdown file, and those same files are what the agent reads. There’s no second copy to drift out of sync. A few agent-only files add depth the pages don’t have room for.

The agent is Claude Haiku with one read-only tool: read a topic. The list of topics sits in its system prompt, so it never spends a round trip asking what exists. It answers from what it read and cites the topic it used. If the answer isn’t in the corpus it says so and points at the contact links — it can’t invent a claim, because it can only see markdown I wrote. What it may say about my employer’s work and a pre-launch startup is fenced off explicitly.

Answers are cached, so repeat questions cost nothing. The starter chips are precomputed at build time, and a per-IP rate limit keeps abuse boring.

Serverless endpoint on Vercel, question log and answer cache in Supabase Postgres.