How I built an AI GitHub repo auditor in one day using Llama 3 + FastAPI

#ai #python #opensource #github

The Problem

Code audits are expensive. Tools like SonarQube cost $500+/month, and most dev teams skip them entirely. I wanted to build something affordable and instant.

What I Built

GitPulse AI analyzes any public GitHub repository in under 60 seconds and gives you:

Overall health score (A-F grade)
Security posture assessment
Technical debt estimate
Team performance metrics
Prioritized AI recommendations

Tech Stack

FastAPI (Python) — backend
Groq API + Llama 3 70B — AI analysis
GitHub REST API — repo data
PostgreSQL — storing results
WeasyPrint — PDF generation
TronScan API — blockchain payment verification

How the AI Analysis Works

I fetch ~15 data points from GitHub API:

Commit frequency and patterns
PR merge rate and open PR count
Contributor distribution
Issue resolution rate
Branch count and release history

All this gets sent as structured context to Llama 3 70B which returns scores across 5 dimensions: code activity, community health, documentation, maintenance, and security posture.

Interesting Technical Challenge

Payment verification without a payment processor. When a user pays in USDT crypto, I verify the transaction directly on the blockchain via TronScan API — no Stripe, no middleman.