Case study · Data
Collection and enrichment for listings and agents across the UAE, France, Spain, Monaco, the UK and the USA.
- 150klistings from one source
- 10kagents from one source
- 70–80%of fields handled by rules alone
The problem
Listing data is messy and protected. Running every field through a language model would be slow, costly and inconsistent.
What we built
- 01
A hybrid extractor: deterministic rules handle most fields, and a local model only classifies the edge cases.
- 02
Waterfall enrichment from public registries, brokerages and aggregators before any paid provider.
- 03
Raw JSON storage, queue-based workers and handling for bot protection.
The result
A single collector proved out at 150,000 listings and 10,000 agents, with model cost and variance kept to the narrow cases that need it.
Next case study
A multi-persona assistant that drives a map