Back to projects03 / Real estate / operator workflow
AI Real Estate CRM
The AI Real Estate CRM is a public systems layer for turning messy property, owner and lead data into explainable matching and safer next actions.
The problem
Property CRMs become noisy quickly: duplicated people, stale property context, weak identity resolution and generic “AI” recommendations make it harder to decide who to call and why.
How I approached it
- 01Model data freshness and evidence instead of treating every CRM field as equally trustworthy.
- 02Use deterministic matching where the operator needs to understand why a contact or property matched.
- 03Generate call angles from verified context instead of inventing buyer or seller intent.
- 04Keep voice-agent handoff separate from identity and matching logic.
03 / WHAT SHIPPED
Key features
- Evidence freshness
- Contact/property identity logic
- Deterministic matching
- Safe call-angle generation
- Voice-agent handoff data
- Synthetic-first public fixtures
Architecture
- Core
- TypeScript
- Domain
- Contacts · properties · leads
- Matching
- Deterministic / evidence-aware
- Voice
- Structured handoff layer
- Fixtures
- Synthetic-first
- License
- MIT
What CRM is built to preserve.
- Never fabricate intent
- Make data freshness visible
- Keep identity resolution auditable
- Design for a broker’s next action
Useful answers, without the hunt.
Does the public CRM repository contain private client data?
The public project is synthetic-first and is designed to demonstrate the systems logic without publishing private operational data.
What makes the matching explainable?
The public core emphasises deterministic matching and evidence freshness so a suggested relationship can be traced to known data.
Is the voice agent the CRM itself?
No. Voice-agent handoff is treated as a downstream workflow; the CRM core first resolves identity, context and the safe call angle.
