Multi-source territorial real-estate scoring
Data pipelineMulti-sourceLLMGCP
~10
heterogeneous sources
7 months
of development
data → decision
complete chain
Problem
Assess the real-estate appeal of a municipality to drive investment decisions, through a complete chain from raw data to decision.
Constraint
About ten heterogeneous sources, including authenticated platforms; ambiguity of homonymous municipalities; a long project run over 7 months.
Approach
- Integration of ~10 sources (INSEE, BPE, Sitadel…) and scrapers for authenticated platforms (Playwright / Selenium).
- Configurable scoring engine: rules externalised in Firestore, editable by administrators.
- LLM analysis producing an investment report; n8n orchestration, GCP / Firestore deployment.
Result
- Complete application in production: geocoding → reference municipality → collection → scoring → report.
- Critical homonymous-municipality bug fixed by propagating an unambiguous EPCI key from geocoding through to the interface.
Stack
Pythonn8nGCPFirestorePlaywrightOpenAI
Next project
CGR International · LLM extraction of technical data for industrial quoting