LeCompliancer

Inspiration

TikTok ships features worldwide, but every region has its own rules (GDPR in Europe, LGPD in Brazil, DSA in the EU, etc.). Today, figuring out if a feature needs special handling is manual and messy. We wanted a tool that flags risks automatically and leaves a clear audit trail.

What it does

LeCompliancer is an AI‑powered checker that:

1. Scans product docs, rollout notes, and code changes.
2. Cross‑checks them with laws and company policies.
3. Flags when a feature needs region‑specific logic (like EU data exports or BR data localization).
4. Creates a Compliance Card with required actions, citations, and a confidence score.

How we built it

1. Inputs: PRDs, design docs, feature flags, and legal texts.
2. LLM + Rules: LLM suggests links between features and laws
3. Output: JSON Compliance Cards with controls and csv, sources, and next steps.

Challenges we ran into

1. Internal jargon (codenames) confused the model: solved with alias mapping.
2. LLM hallucinations: solved by forcing every answer to include a legal citation.
3. Balancing false alarms vs. missed risks.

Accomplishments that we're proud of

1. Working prototype that flagged >80% of known compliance needs correctly.
2. Near‑real‑time: PRD: Compliance Card in minutes.
3. Clear audit log of every decision.

What we learned

1. LLMs work best when combined with simple rules.
2. Structure and citations beat free‑text by using prompt engineering.

What's next for LeCompliancer

1. Improve precision by doing more testing, training and refining prompt
2. More automation: agents that re‑check features when laws change.
3. One‑click compliance briefs for regulators.
4. Scale dashboards so PMs, lawyers, and engineers can all see the same truth.

Built With

• craw4ai
• fastapi
• gemini
• json
• langchain
• pydantic
• python
• react
• yaml