Turn messy WhatsApp broadcasts into structured listings
Property agents send free-form, bilingual listing messages. This demo runs each one through an LLM pipeline that parses, validates, and structures it, using confidence-based model escalation and correctness invariants. It runs on a deterministic fake model, no API key required, so every run gives the same result.
Parser playground →
Paste a broadcast and watch it become structured fields, with the model tier and confidence that produced it.
Listings →
The listings parsed so far this session, served by the NestJS REST API. This is an in-memory demo store — parsed records reset when the backend restarts.