Compass
An AI-native compliance operating system that combines deterministic rules with LLM reasoning — with an auditable decision log and an explicit human approval boundary for high-risk content.
Problem
Compliance reviews in regulated industries are slow, error-prone, and nearly impossible to audit consistently.
Plugging an LLM into the process without accountability boundaries creates legal and reputational risk.
Teams need a structured workflow with clear ownership — not a chatbot that might be wrong.
Solution
Built a full-stack compliance workflow: content comes in, gets risk-classified, receives a compliant rewrite suggestion, and has required disclosures attached.
A deterministic rules layer handles predictable checks; LLM reasoning is invoked only where judgment is genuinely needed.
High-risk content is gated behind an explicit human approval step, with every decision persisted to an auditable log.
Key Engineering Decisions
Designed accountability into the architecture: AI assists and recommends, humans approve — the boundary is explicit and enforced.
Persisted structured decision records (inputs, outputs, model rationale) so every review is reproducible and defensible.
Kept rules, models, and UI as independent modules so each can evolve without breaking the others.
What It Demonstrates
AI workflow redesign with real accountability — not just LLM integration bolted onto an existing process.
Risk-based gating and full traceability designed to map to actual compliance and legal requirements.
Full-stack ownership from FastAPI backend and SQLite persistence through to React + Next.js frontend.