01
Contact verification
Confirms the entity's real, reachable footprint.
Five agents, one verdict.
The client vetted companies by hand — Google searches, review sites, news scans, legal history. Hours per company, every time. We replaced the process with a supervised multi-agent system that returns a sourced risk report in minutes.
The five agents
02
Live news scan
Pulls current coverage and incident signals.
03
Review analysis
Reads sentiment and patterns across review sites.
04
Legal history
Surfaces filings, disputes, and enforcement records.
05
Legitimacy assessment
Weighs registration and operating evidence.
Parallel, not sequential
One AI walking a checklist is bound by the sum of its steps. Five agents each own a single job and run at the same time — total runtime is the slowest agent, not the total.
Live data only
Every agent queries Google Search through Serper at request time. Nothing cached, nothing stale — the report reflects the company as it is today.
Validated before it counts
Agent output passes JSON schema validation before it reaches the report. Every claim has to be structured and sourced, so nothing unverifiable makes it through.
Fails loud, not silent
Error handling sits at every stage. If a live search comes back empty, the run flags the gap and keeps going — no crash, and no silent hole in the report.
A human signs off
The scored report lands in Google Sheets with its evidence trail attached. A reviewer checks it before it reaches the client. The system does the research; a person still makes the call.
5
Agents in parallel
0–7
Risk scoring scale
Hrs → min
Research turnaround
Research that used to take hours now runs in minutes — with a paper trail showing exactly how every score was reached.