FollowMe

Automated GitHub repo discovery, NVIDIA NIM LLM grading, and auto follow/star tool

Dashboard

How It Works

1. Discovery: Scheduled job searches GitHub for active repositories created in the last 7 days matching targeted topic tags.
2. AI Evaluation: Fetches the README snippet and submits it to NVIDIA NIM (meta/llama-3.1-8b-instruct), grading the repository from 1 to 10 with a focus on student learning effort, original prototypes, and community builders.
3. Smart Follow Filter: Filters out "ego" developer accounts. Follows are executed only if the target user has a peer-profile signature (20-500 followers, following > 20, ratio 0.5-2.0, account age > 6 months). High-profile accounts are starred but skipped for follows.
4. Data Sync: Stores evaluation history, grades, actions, skip logs, and follow statuses in Supabase.
5. Periodic Cleanup: Runs every 6 hours checking all auto-followed accounts. If they fail to follow back within 3 days, it automatically unfollows them to maintain healthy account statistics.

Targeting Guidelines

Follows are filtered through active peer criteria:

• Primary Targeting: followers: 20–500, following > 20, followers/following: 0.5–2.0, and age > 6 months.
• Celebrity Bypass (Skip Follow / Star Only): Triggered if followers > 500 and following < 10.

Architecture

• Worker: Node.js + Express + node-cron on Render
• Grading Engine: NVIDIA NIM LLM Integration
• Storage Layer: Supabase PostgreSQL database
• Web Console: Next.js 15 UI with monochrome design dashboard on Vercel

Live Endpoints

• Dashboard: https://followme-mauve.vercel.app
• Worker status: https://followme-gg6q.onrender.com/health