ClarityAgent

Inspiration

I vibe-coded interceptor.cpp in my SentinelAPI project and got completely called out in code review. I couldn't explain a single line. That's the problem ClarityAgent solves.

What it does

ClarityAgent connects to any GitLab repository, scans every source file with Gemini AI, scores each file by how well the author truly understands it, and generates personalized study plans for high-risk files — before an interviewer exposes the gaps.

How we built it

FastAPI backend fetches file trees via GitLab API. Gemini Flash analyzes each file and scores understanding depth 0-100. MongoDB Atlas stores all analysis results and session history. Vanilla JS frontend displays the Understanding Map dashboard.

Challenges we ran into

Gemini free tier rate limits when analyzing multiple files simultaneously. GitLab API pagination for large repositories. Building a responsive real-time analysis pipeline.

Accomplishments that we're proud of

A working agent that actually detects vibe-coded files. Professional dashboard showing per-file risk scores. Full GitLab + MongoDB + Gemini integration in one night.

What we learned

How to integrate GitLab's REST API, structure agentic pipelines with Gemini, and use MongoDB Atlas for storing AI analysis results.

What's next for ClarityAgent

GitHub support, auto-creating GitLab issues with study plans, team-level understanding dashboards.

Built With

• fastapi
• gemini-ai
• gitlab-api
• html/css
• javascript
• mongodb-atlas
• python