Skip to content

README.md

Latest commit

 

History

History
122 lines (86 loc) · 4.19 KB

README.md

File metadata and controls

122 lines (86 loc) · 4.19 KB

🕵️‍♂️ AI GitHub Auditor (Forensic Edition)

"Code that isn't documented doesn't exist."

An opinionated, forensic audit tool that grades GitHub profiles and repositories for job-readiness. It separates Signal from Noise using strict engineering standards, detecting tutorial clones, low-effort forks, and "marketing fluff" that lacks code evidence.

License Status Stack

🔴 Live Demo | ⚙️ API

🚨 The Problem

Most GitHub profiles fail not because developers lack skill, but because their best work is buried. Recruiters spend <10 seconds on a profile. If they see:

  • ❌ 30+ "Tutorial" repos (e.g., react-course-chapter-1)
  • ❌ Forks with zero commits
  • ❌ "Full-Stack" apps with empty READMEs
  • ❌ "AI Projects" that are just API wrappers

...they close the tab. You lose the interview.

🎯 What This Tool Does

This is not a "nice" AI. It acts as a Cynical Senior Hiring Manager. It audits your profile in two stages:

1. The "First Impression" (Profile Audit)

Before looking at code, it judges your Presentation:

  • Does your Bio & README clearly state your value?
  • Do you have a professional avatar and links?
  • Verdict: Assigns a Profile Score (0-100) and an RPG-style Class (e.g., "The Full-Stack Catalyst" or "The Ghost Developer").

2. The "Forensic Deep Dive" (Repo Audit)

When you ask it to verify a specific repository, it performs a Lethal Audit:

  • Boilerplate Detection: If >50% of your code is config/JSON, you get a penalty.
  • The "Ghost" Rule: Great code with no README = Max Score 75.
  • One-Day Wonders: If you built a "complex app" in <24 hours, it flags it as a "Rush Job" (Max Score 60).
  • Claim Verification: If you claim "AI-Powered" but have no ML libraries, it flags a "Marketing Mismatch".

📊 The Scoring System (Strict Mode)

We use Llama 3.3 (via Groq) with a low temperature to ensure objective grading.

Tier Score Definition
🏆 FLAGSHIP 90-100 Real-world app. Complex logic (Auth, State, CI/CD). MUST have extensive documentation.
🛠️ SOLID 70-89 Clean code that works. Caps at 80 if it's a generic "Buzzword Stack" (MERN CRUD) or lacks a README.
😐 NEUTRAL 40-69 Config files, simple scripts, "Profile READMEs", or repos with >50% boilerplate.
🗑️ NOISE 0-39 Forks, "Hello World" tests, empty folders, or obvious tutorial clones (scrimba, bootcamp).

🧱 Tech Stack

  • Frontend: React, Tailwind CSS (Dark Mode UI)
  • Backend: Node.js, Express, TypeScript
  • AI Engine: Groq SDK (Llama-3.3-70b-Versatile)
  • Data Source: GitHub REST API

🚀 Getting Started

Prerequisites

  • Node.js (v18+)
  • A free Groq API Key
  • (Optional) GitHub Personal Access Token (for higher rate limits)

Installation

  1. Clone the repo

    git clone [https://github.com/prem22k/ai-github-auditor.git](https://github.com/prem22k/ai-github-auditor.git)
cd ai-github-auditor

  2. Setup Backend

    cd backend
npm install
# Create .env file
echo "GROQ_API_KEY=your_key_here" > .env
npm run dev

  3. Setup Frontend

    cd frontend
npm install
npm run dev

  4. Audit Yourself Open http://localhost:5173 and enter your GitHub username.

⚠️ Disclaimer

This tool is opinionated. It mimics the harsh reality of tech hiring.

  • It will hurt your feelings if your repos are empty.
  • It will give you a low score if you pushed a repo 5 minutes ago (The "One-Day Build" penalty).

The goal is not to validate you. The goal is to get you hired.

👤 Author

Prem Sai Kota

"Building tools that tell the truth."