Your door to Japan's job market.
A full-stack AI job search assistant built for Japan-focused engineers. Paste a job URL, get an instant resume match score, and track your pipeline — all in a dark, minimal UI powered by Gemini.
Pipeline (Kanban) Drag jobs across four stages — Unsorted, Interested, Applied, Pass. Each card shows an AI fit score as an animated ring. Bulk-score all unscored jobs with one click; a 4-second rate-limit delay prevents Gemini quota burn.
Resume Management Upload multiple PDF versions (e.g. "Backend v2", "Full-stack"). Each is parsed by Gemini into structured JSON: skills, years of experience, work history. Set one as active — it becomes the baseline for all scoring.
Job Parsing Paste any public job listing URL (104, CakeResume, Japan Dev, Tokyo Dev, etc.). Gemini fetches and extracts title, company, required skills, salary, location, and remote policy.
Match Analysis Semantic comparison between your active resume and any parsed job. Returns a score 1–10, matched skills, missing skills, and concrete suggestions. Cover letter generation (Traditional Chinese + English) with adjustable tone.
Application Tracking
Track every application through pending → applied → interviewing → offer / rejected with free-text notes per stage.
| Layer | Choice |
|---|---|
| Frontend | Next.js · TypeScript · Tailwind CSS · Framer Motion |
| Drag & Drop | @dnd-kit/core |
| Backend | FastAPI · SQLAlchemy 2.0 (async) |
| Database | PostgreSQL |
| Vector DB | ChromaDB |
| LLM | Gemini API (free tier) |
| PDF Parsing | pdfplumber |
| Package Manager | uv (backend) · npm (frontend) |
| Deployment | Docker Compose |
- Docker
- Node.js 18+ (for the frontend)
- uv — Python package manager
- A Gemini API key (free tier works)
git clone https://github.com/your-username/tobira.git
cd tobira
cp .env.example .envEdit .env:
DATABASE_URL=postgresql+asyncpg://fitcheck:fitcheck@localhost:5432/fitcheck
CHROMA_HOST=localhost
CHROMA_PORT=8001
GEMINI_API_KEY=your_key_here
# Optional: Crawler Jobs page (Japan Dev + Tokyo Dev daily listings)
# Get these from your Supabase project → Settings → API
SUPABASE_URL=https://[PROJECT_REF].supabase.co
SUPABASE_KEY=your_supabase_anon_key_heredocker compose up --buildThis starts PostgreSQL, ChromaDB, and the FastAPI backend with hot-reload on port 8000.
cd frontend
npm install
npm run devOpen http://localhost:3000.
| URL | Description |
|---|---|
http://localhost:3000 |
Tobira UI |
http://localhost:8000/docs |
FastAPI Swagger docs |
tobira/
├── docker-compose.yml
├── .env.example
├── frontend/ # Next.js app (App Router)
│ └── src/
│ ├── app/ # Pages: /, /resume, /jobs, /match, /apps
│ ├── components/
│ │ ├── kanban/ # Board, JobCard, ScoreRing
│ │ └── layout/ # Sidebar
│ └── lib/
│ ├── api.ts # Typed API client
│ └── types.ts # Shared TypeScript interfaces
└── backend/
├── Dockerfile
├── pyproject.toml
├── tests/ # pytest suite (100% coverage on core modules)
└── app/
├── main.py # FastAPI app + CORS
├── core/
│ ├── gemini.py # Gemini API client
│ ├── vector_db.py # ChromaDB client
│ └── supabase_db.py # Supabase PostgREST client
├── routers/ # API endpoints
├── services/
│ ├── parser.py # PDF + resume parsing
│ ├── scraper.py # Job URL fetching via Gemini
│ ├── embedder.py # Embedding generation + storage
│ ├── matcher.py # Resume ↔ job analysis
│ └── generator.py # Cover letter generation
└── models/ · schemas/
- LinkedIn is not supported (requires authentication)
- Gemini free tier: ~20 generation requests/day, 1000 embedding requests/day
- Resume and job data is sent to the Gemini API (Google's terms apply)
- The Pipeline page's crawler jobs require a running jp_job_crawler Supabase project — scores are cached locally so each job only consumes one Gemini call