Tomb rAIder: AI Search with Kotlin

Most teams build AI search one of two ways: embed everything and run vector similarity, or hand the whole query to an LLM. Both hit limits — embeddings can’t enforce hard requirements, and a single large prompt is slow, brittle, and hard to test.

This talk shows a more deliberate approach. Building a real search engine in Kotlin — running fully locally with Spring AI, Postgres + pgvector, and LM Studio. I show how to decompose a query and route each part to the tool that’s actually best for it.