Multi-Agent Ethical Document Reviewer

Ethical Agents is an intelligent document review platform that automatically screens any document — policy, AI system design, research proposal, or product specification — for ethical risk before publication. The system runs five specialist AI agents in parallel, each responsible for a different risk domain: Bias, Privacy, Security, Compliance, and Transparency. Every agent evaluates the document against a curated Domain Knowledge Repository of risk fields using a four-component scoring model that measures keyword coverage, term density, and topical prominence. A governance multiplier then adjusts scores based on the document's ethical posture — rewarding explicit safeguards like human oversight and audit trails, and penalising dismissals of safety mechanisms. A parent orchestrator aggregates the five domain scores using a ceiling-pull model, ensuring a single severely problematic domain cannot be outweighed by clean ones. The final output includes a clear Publish or Do Not Publish decision, an overall risk score out of 100, per-domain breakdowns, a list of every high-risk field with the reason it was flagged, and concrete remediation steps. The system also features a self-improving learning loop: when connected to a Fireworks AI LLM, agents can identify novel risk concepts not yet in their knowledge base and permanently extend it. A calibration pipeline with automated grid search keeps scoring accuracy tuned against labelled test documents. The React frontend displays all results in real time and loads the full analysis history from the backend on startup.