Multi-agent compliance SaaS platform for AI systems
- 7 minper assessment, was days
- 50+real analyses and pilots
- 19persistent agents
- 193guidance measures covered
The problem
Assessing whether an AI system complies with European and national regulatory frameworks took days of manual review: reading the regulation, checking evidence and cross-validating across 12 dimensions, 193 guidance measures and 16 official documents.
And it ran on declarative questionnaires: the result depended on what someone claimed, not on what the system does.
The architecture
A B2B SaaS platform on Azure AI Foundry with 19 persistent agents specialised in compliance dimensions, legal analysis, scoring and traceability.
It analyses the project’s real artefacts — source code, documentation, models and technical evidence — across roughly 510 Section × Measure pairs. Each dimension emits a score, detected evidence, prioritised gaps and the regulatory reference behind them.
Impact
- From days of manual analysis down to 7 minutes per assessment
- Over 50 real analyses and pilots run
- 12 dimensions and 193 guidance measures across some 510 assessment pairs
- Assessment grounded in the project's real evidence, not in forms
Stack
- Azure AI Foundry
- Multi-Agent
- Python
- FastAPI
- React
- Scoring Engine