An auto-curating map of the open-source AI jungle

What it does

Hello-AI is an automated curation engine that scans GitHub for AI projects with at least 500 stars, feeds them in batches to an LLM for tagging and categorization, and renders the results into a VitePress directory. It currently tracks 19,493 collected projects, though only 9,539 active ones—those updated within the last six months—make it to the live site. The system also auto-purges stale entries and maintains an objective “Trending” list based purely on recent star velocity, overriding the AI’s taste for that category.

The interesting bit

The pipeline treats the LLM as a bureaucrat: it dynamically reads existing categories from projects.json, instructs the model to route new repos accordingly, and quarantines rejects into an audit trash folder. Because the categories themselves are not hard-coded, the taxonomy evolves as the crawler discovers new topics, which are then seeded back into the exploration queue. It is essentially glue code orchestrating GitHub’s API, a generic OpenAI-compatible LLM, and a static site generator into a closed loop.

Key highlights

• Discovers repos via GitHub API using a seed topic list that auto-expands when the crawler spots new keywords
• Evaluates projects in concurrent LLM batches to save tokens and respect rate limits
• Supports local and remote LLMs (OpenAI, DeepSeek, MiniMax, or Ollama via local-fallback)
• Frontend navigation and sidebars regenerate dynamically from the database without manual mapping
• Rejected projects land in data/rejected-projects/ for audit rather than silent deletion

Caveats

• The “24/7 autonomous” maintenance requires running local daemon scripts like scripts/loop-eval.js; it is not a managed service
• The README is upfront that the project is “entirely” automated scripts and LLM collaboration, so expect occasional LLM hallucinations in categorization

Verdict Worth a look if you want a self-hosted discovery engine for AI repositories or need inspiration for automated curation pipelines. Skip it if you are just looking for a static list—this is a framework that requires an LLM API key and a GitHub token to actually function.