Natural-language search
Accept a project brief and stream progress through understanding → ranking.
NeedBuilders think in outcomes, not GitHub query syntax.
ValueOne gesture from intent to candidates.
ProjectsCase Study
GitSearch AI
Describe what you want to build—AI searches GitHub, reads READMEs, and returns ranked repos with scores, strengths, and gaps.
- Industry
- Developer tools · Open-source discovery
- Category
- AI product · GitHub search
- Role
- Full Stack Product Engineer — product, search quality, and UI end to end
- Timeline
- 2026
- Status
- MVP product build
- Stack
- Next.js · TypeScript · Tailwind · Gemini / OpenAI · GitHub REST · NDJSON streaming
Executive Overview
GitSearch AI replaces manual GitHub keyword digging. Paste a build brief; get a short scored list of repos with strengths and gaps—so you decide, not scroll.
The product surface stays thin on purpose: landing + search, no auth, no database. The hard work is ranking quality under cost and latency limits, plus a wait that stays understandable.
Business Problem
Existing workflow
Developers guess GitHub keywords, skim awesome-lists, and open README tabs until something looks close enough.
Pain points
- A full project brief does not map cleanly to GitHub’s keyword search.
- Popular curated lists crowd out repos you can actually build on.
- Long silent AI requests feel broken even when work is still running.
Why it matters
The wrong starter wastes days. A ranked shortlist turns discovery into a decision.
Solution
One product loop:
Brief in
Describe what you want to build in plain language on the landing or /search.
Ranked shortlist out
See a top-five list with scores, strengths, gaps, and tech tags—while progress streams live.
Faster open-source discovery with a decision-ready shortlist, not a dump of GitHub links.
Core Features
Discovery
Ranked repo shortlist
Top matches with score, strengths, gaps, and tech—plus clear retry when upstream APIs fail.
NeedTrust comes from seeing why a repo ranked.
ValueA shortlist you can act on.
Technical Highlights
Ranking under hard caps
Brief → short keywords → multi-strategy GitHub search → demote awesome-list style repos → README enrichment → batched AI scoring with candidate and concurrency limits.
Business valueUseful rankings inside a fixed latency and token budget.
Progress you can trust
POST /api/search streams NDJSON steps so the UI shows work in motion instead of a silent 60s spinner.
Business valuePeople stay with the search long enough to get a result.
Challenges & Solutions
| Challenge | Solution |
|---|---|
| Plain-language briefs produced noisy GitHub queries, and curated “awesome” lists ranked like buildable starters. | Short keyword extraction, staged search strategies, and demotion rules (heuristics + scoring) that push list repos down. |
| Scoring every search hit with full READMEs blew past latency and token budgets. | Heuristic shortlist first, truncate READMEs, then evaluate in capped batches with concurrency limits. |
Tech Stack
- Frontend
- Next.js App Router, React, TypeScript, Tailwind
- API
- POST /api/search — NDJSON stream, 60s budget
- AI
- Gemini or OpenAI — keyword extraction and repo scoring
- External
- GitHub REST Search + README fetch
Need a focused AI tool that ships one job well?
GitSearch AI is a thin MVP case study: one discovery loop, real ranking constraints, and a clear wait—not a platform story. If you want similar craft on a developer tool, start a conversation.
Start a conversationRole
Full Stack Product Engineer
Client
Personal product / AI developer tool
Date
2026
Collaborators
Solo build
Next Project