The best rooms never reach Airbnb.
I kept finding them in local Telegram groups instead — buried in the scroll, gone within hours. So I built Koloq: it reads those groups and turns messy posts into a searchable feed and map — price, dates, neighborhood. Nothing is hosted here; from every card you talk to the author directly in Telegram.
12 Telegram groups · 9 countries · ~70 new posts a day
A Telegram parser pulls in new posts daily. An LLM pipeline — model routing, an embedding pre-filter, cost controls — extracts price, dates, housing type and location from free-form text. The feed ranks by freshness and stay horizon. TypeScript end-to-end: Node + Postgres, React, self-hosted, with OpenTelemetry and Grafana for observability.
Koloq is a solo project — parsing, LLM pipeline, product, design and infra by one engineer. I'm open to work: applied AI / GenAI engineering — LLM pipelines, evals and the product around them.