A compliance program run by agents, checked by people

about project
Alchemer · 2025 to present · Global Compliance Manager
Problem. One compliance manager, seven frameworks (SOC 2 Type 2, ISO 27001, GDPR, HIPAA, CCPA, FERPA, EU-US DPF), and hundreds of customer security assessments a year. Answering from memory is slow and drifts out of date.
What I did.
Built a governed knowledge base, one maintained page per topic, so every answer traces to a source.
Agents draft assessment answers only where the knowledge base supports them and leave the rest blank. A separate read-only agent re-checks every flagged item. I review and submit.
Every answer I correct is logged with its cause (source wrong, source stale, wrong page, misread), and the source is fixed so the error doesn’t come back.
Wrote the company’s Responsible AI program on ISO/IEC 42001: policy, risk and impact assessments, systems register, model cards. Eight documents, approved by the CISO.
Put the tooling under a CI gate of 2,600+ automated tests that blocks any change that fails, including my own.
Result. Customer security assessment turnaround went from 5 days to 2 hours, with every answer traced to a source and reviewed by a person before it goes out.