Willi
Willi lives inside Telegram, where rental discovery is fast, informal, and noisy. To feel native, search had to be near-instant and results had to be trustworthy. The challenge was obvious: dozens of channels and chats, inconsistent formats, reposts, and expired listings—great for humans to browse, terrible for machines to search.
We engineered a Telegram-aware crawler that continuously gathers posts and stories, normalizes content, and extracts the fields renters actually use—location hints, price, term, contact, and basic amenities. A de-dup and freshness layer prunes reposts and flags likely stale ads; idempotent ingestion and rate-limit-aware workers keep the pipeline quiet even during spikes.
On the serving side, we optimized for “first useful result” speed. Listings are indexed with lightweight embeddings and structured facets, so users can filter by area and budget while still catching near-matches. Caches warm along popular routes, relevance boosts recent and verified entries, and graceful fallbacks ensure something helpful appears even when upstream sources wobble.
For the business, this translated into calmer operations and happier users. Hosts saw steady, qualified inquiries instead of bursts of noise. Renters found live options faster and returned more often because the inventory felt fresh and credible. The story isn’t flashy: it’s disciplined crawling, clean data, and a search experience that simply works where people already are.
