AI-native Compliance Operating System
Compass reimagines how regulated content is reviewed by rebuilding compliance workflows as AI-native systems rather than layering AI onto legacy processes.
Instead of relying on manual checklists and static documentation, Compass combines a deterministic rules engine with structured LLM reasoning to classify regulatory risk, generate compliant rewrites, attach required disclosures, and persist auditable human decisions.
This project was built as a demonstration of end-to-end AI-assisted workflow redesign in regulated environments.
Traditional compliance review workflows evolved before modern AI existed. Compass redesigns the workflow from scratch:
- AI assumes cognitive responsibility for classification and drafting.
- Humans retain legal accountability for release decisions.
- Every assessment and decision is logged for auditability.
- Lifecycle state is visible at a glance.
This is not an AI wrapper. It is an AI-native operating model.
- FastAPI for API layer
- SQLite for persistence and audit logging
- Deterministic rules engine for policy evaluation
- Optional LLM augmentation via OpenAI API
- Structured JSON responses for explainability
- Next.js + React
- Tailwind CSS for clean operator-focused UI
- Lifecycle-aware sidebar with decision status indicators
- Explicit human review panel with persisted decisions
- Risk classification (LOW / MEDIUM / HIGH)
- Rules-first policy pack architecture
- LLM-assisted reasoning and rewrite generation
- Required disclosure generation
- Explicit human stop boundary
- Persisted audit trail of:
- Assessments
- Risk tags
- Rewrites
- Human approval / rejection decisions
- Decision lifecycle visualization
- User submits regulated content.
- System evaluates via deterministic rules.
- Optional LLM augments reasoning and rewrite suggestions.
- High-risk assessments require human review.
- Human decision is logged with reviewer + notes.
- Lifecycle status updates in real time.
- Regulatory drift
- Policy version skew
- Inconsistent interpretation across jurisdictions
Compass is structured to support policy pack versioning and clear decision traceability to mitigate these risks.
Backend:
- Python
- FastAPI
- SQLite
- OpenAI API
Frontend:
- Next.js
- React
- Tailwind CSS
Responsible AI deployment in regulated environments requires:
- Clear accountability boundaries
- Deterministic fallbacks
- Explainability
- Auditability
- Human oversight
Compass demonstrates a practical implementation of those principles.