GitHub's issue search sucks. So I vibe-coded my own.

GitHub's issue search is broken. So I built my own.

You type "good first issue Python" and get 50,000 results. Half are from 2019. A quarter are already assigned. The rest? Good luck figuring out which repos are actually active.

So I did what any reasonable developer would do: I opened Antigravity, told Claude what I wanted, and vibe-coded the whole thing over a weekend.

The result: GitHub Contribution Finder - a search engine that actually understands what you're looking for.

What it does

Type something like:

"Python issues in ML projects updated this week"
"Unassigned TypeScript bugs in popular repos"
"Documentation fixes I can do in an hour"

The AI parses your intent, searches a vector database of 50,000+ issues, and returns results you can sort by newest, recently discussed, AI relevance, or stars.

The stack (for the nerds)

Frontend: Next.js 15, TypeScript, Tailwind, shadcn/ui
Backend: FastAPI (Python)
AI: Google Gemini (query parsing + embeddings)
Vector DB: Pinecone
Data Ingestion: GitHub GraphQL API + GitHub App (15,000+ requests/hour)

How I actually built this (the vibe-coding part)

I'll be honest: I didn't write most of this code by hand. Here's what the process actually looked like:

Friday night:

"I want a search engine that uses AI to find GitHub issues." Antigravity + Claude Opus 4.5 generated the FastAPI backend and GraphQL fetcher. I tweaked the Pinecone integration and went to bed.

Saturday:

Built the frontend. Described the UI I wanted, accepted most suggestions, rejected a few. Added filters for language, labels, and time ranges. The hardest part was getting the ranking weights right. Ended up at 70% semantic relevance, 30% recency.

Sunday:

Polished the UI, added the "Last Updated" indicator so users know the data is fresh, and deployed to Vercel + Cloud Run.

Total time: ~15 hours across 3 days. Is the code perfect? No. Does it work? Yes.

The features I'm most proud of

1. Semantic search (not just keywords)

If you search "beginner," you won't get issues that say "this is NOT for beginners." The AI understands context.

2. Time filters

Filter issues from the last 2 hours, 24 hours, 7 days, or 30 days. No more stale issues from 2020.

3. Smart sorting

Newest: By creation date
Recently Discussed: By last comment
Relevance: AI-ranked
Stars: For the popular repos

4. Data freshness indicator

Right on the homepage, you can see when the database was last synced. Transparency.

What I learned

Vibe-coding is real. I wouldn't have built this in a weekend without AI assistance. Maybe not even in a month.
GitHub's API is brutal. Rate limits, abuse detection, secondary limits. I had to upgrade from a Personal Access Token to a GitHub App just to get enough requests.
Ranking is harder than search. Getting the weights right (relevance vs. recency vs. popularity) took more time than building the actual search.