Trouve·Ka

Trouve·Ka is a real search engine — not a metasearch. Crawler, frontier, parser, Quebec classification, index, ranking, semantics: everything is built in-house, with no dependency on Google, Bing or Brave for results. Born as an open Quebec-web engine (64 high-authority seeds, 7,898 discovered domains), it pivoted on 2026-08-23 to become the search engine of Groupe Ka: it continuously indexes the ecosystem's 14 sites, fed by daily sitemap ingestion — the frontier alone exceeds 4.6M pending URLs.

The engine's motto: crawl continuously, index immediately, search immediately, improve asynchronously. Every fetched page is scored deterministically for Quebec relevance (TLDs, toponyms, postal codes, area codes, known organizations, language — no LLM in the hot path) and indexed inline in OpenSearch: searchable in about 2–4 seconds. Enrichment — domain authority, entities, embeddings, reranking — arrives afterwards via Redis Streams without ever blocking, with 100% semantic coverage of the index.

In production: 302,000+ pages indexed (and growing by 100K+ on good days), bilingual BM25 ranking with FR/EN synonyms (thermopompe ↔ heat pump), semantic search plus a dedicated reranker, and 258–425 ms measured latency with both active. The crawl is polite and self-healing: an identified TrouveKABot honoring robots.txt and crawl delays, SSRF guards, crawl-trap detection, and nine workers across five cluster nodes coordinated without an orchestrator through Postgres FOR UPDATE SKIP LOCKED — proven to survive node reboots on its own.