An acquisition-sourcing tool that scores and prioritises companies deterministically, and keeps the LLM strictly on the explaining side of the line.
The problem
Lead platforms return hundreds of companies. The buyer still has to decide which deserve enrichment credits and a phone call — and be able to defend that decision later. An opaque AI score cannot be defended.
The target list, running locally. The synthetic-data badge is part of the app, not added for this page.
A prototype running on synthetic data — fictional companies and .example domains. It does not perform enrichment itself, and there is no hosted demo.
What I built
Two scores kept deliberately separate: how well a company fits the buy box, and whether the record is complete enough to act on. A good target with a thin record is a different problem from a bad target.
Scoring is rule-based and reproducible — six weighted factors, every point traceable to a rule, the same input always producing the same score. The drawer shows the arithmetic.
The LLM reads the scored record and writes the reasoning, the missing-information summary and the suggested next action. It never calculates or changes a score.
Company identity is a SHA-256 of the normalised domain rather than the name, so 'Acme Inc' and 'ACME LLC' converge; contacts deduplicate on normalised email.
Any scores present in an imported CSV are discarded and recomputed, so there is exactly one scoring engine.
