goodFirstHuman

Finding your first open source contribution shouldn't feel like searching for a needle in a haystack. “Good first issue” labels exist on thousands of repositories, but they tell you nothing about whether you are the right person for that issue.

The Idea

goodFirstHuman flips the model.

Instead of asking contributors to browse and filter endlessly, it builds a profile of who they are:

• Tech stack

• Experience level

• Domains of interest

• Preferred type of work

Using this, it surfaces issues that genuinely fit them.

How It Works

Under the hood, a continuous pipeline:

• Ingests issues from GitHub

• Enriches each issue with:

  • Difficulty scoring

  • Tech stack detection

  • Repository health metrics

  • Mentorship availability flags

• A 1024-dimensional semantic embedding (via HuggingFace BGE model)

All of this is stored in PostgreSQL with pgvector.

At query time, a three-tier ranking system is applied:

1. SQL filtering

2. Cosine similarity (semantic matching)

3. Weighted domain + repository health scoring

This produces a ranked list of personalized recommendations, along with transparent match explanations shown directly in the UI.

The Result

Beginners see:

• Which repositories want someone like them

• Why an issue matches their skills

• How approachable the community is

Experienced contributors get:

• Issues that challenge them at the right level

All before they even click a link.

Tech Stack

• Next.js

• Node.js

• PostgreSQL

• pgvector

• Meilisearch

• Redis

• BullMQ