GitHire

Landing Page

Inspiration

Recruiters spend hours scrolling through resumes and half-read GitHub profiles, yet still wonder who can truly ship production-ready code.
We imagined an “instant tech recruiter” that evaluates a developer the way engineers do: by reading their commits, not their buzzwords.

What it does

GitHire automatically:

Scrapes + scores code quality
Shallow clones each repo, samples representative files, and runs a custom readability/complexity rubric
Extracts skills & experience
Parses repo metadata, READMEs, topics, and commit messages to infer stacks
Matches candidates to roles
Compares extracted skills to a job spec in real-time, returning a weighted match %, matched repositories, and an AI-generated rationale.
Visualizes everything in a Next.js dashboard so recruiters can sort, search, and filter their ideal candidate.

How we built it

Backend:
- Python / Flask REST API
- PyGithub for metadata, GitPython (depth = 1) for shallow clones
- Lizard CLI for cyclomatic complexity
- Google Gemini 1.5-Flash for natural-language scoring & summaries
- Redis in-memory cache to avoid re-cloning the same repo twice
Frontend:
- Next.js 14 App Router
- Tailwind CSS + shadcn/ui components
- Dashboard pages driven by SWR hooks hitting the Flask API
DevOps:
- Docker-compose for one-command spin-up
- GitHub Actions CI (lint, unit tests, type-check)

Accomplishments that we're proud of

Built an end-to-end pipeline (clone → score → match → visualize) in < 36 hours.
Achieved consistent 95 %+ JSON compliance from the LLM after prompt-engineering.

What we learned

Shallow git clones + smart sampling beat brute-force every time.
Cyclomatic complexity alone ≠ code quality; indentation consistency & comment ratio correlate surprisingly well with readability.
Prompt rigidity (explicit JSON schema, no prose) is the key to reliable LLM integration.