CareerLens

AI-powered resume vs job description analyzer. Paste your resume once, then check your fit against any job description — get a match score, matched skills, gaps, and personalized positioning advice. Powered by a RAG pipeline built from scratch.

🔗 Live app: career-lens-tan.vercel.app 🔗 API docs: careerlens-a8jl.onrender.com/docs

---

What it does

1. Paste your resume — it's chunked and embedded into a vector database
2. Paste any job description — CareerLens retrieves the most relevant parts of your resume using semantic search
3. An LLM analyzes the match and returns a structured breakdown: match score, skills you have, skills you're missing, and how to position yourself for the role

No generic "chat with your PDF" — this is a focused tool solving a real, specific problem with a complete RAG pipeline underneath.

---

Architecture

┌─────────────────┐         ┌──────────────────┐         ┌─────────────────┐
│   Next.js 16     │  HTTP   │   FastAPI         │  SQL    │   Supabase       │
│   (Vercel)       │ ──────► │   (Render)        │ ──────► │   PostgreSQL     │
│                  │         │                   │         │   + pgvector     │
│  Server Actions  │         │  /resume/         │         │                  │
│  ScoreRing       │         │  /analyze/        │         └─────────────────┘
│  SkillChips      │         │  /health          │                  ▲
└─────────────────┘         └─────────┬─────────┘                  │
                                       │                             │
                                       ▼                             │
                              ┌──────────────────┐                  │
                              │  Google Gemini    │                  │
                              │  embeddings +     │──────────────────┘
                              │  generation       │   (vector storage)
                              └──────────────────┘

Resume ingestion (POST /resume/):

raw text → chunked by paragraph → each chunk embedded (Gemini) → stored in pgvector

JD analysis (POST /analyze/):

JD text → embedded → cosine similarity search against stored chunks
        → top 5 relevant chunks retrieved → sent to Gemini with structured prompt
        → JSON response: match_score, matched_skills, missing_skills, positioning_advice

---

Tech stack

Backend

• FastAPI — async-ready Python web framework
• SQLAlchemy + Supabase (PostgreSQL) — data layer
• pgvector — vector similarity search, no separate vector DB needed
• Google Gemini (gemini-embedding-001, gemini-2.5-flash) — embeddings + generation
• Pydantic — request/response validation, settings management
• pytest — test coverage on core pipeline logic
• uv — dependency management

Frontend

• Next.js 16 (App Router, Server Actions)
• React 19 (useActionState)
• TypeScript
• Tailwind CSS
• Custom animated SVG score visualization

Infrastructure

• Render — backend hosting
• Vercel — frontend hosting
• Supabase — managed PostgreSQL + pgvector
• GitHub Actions–ready monorepo structure

---

Why these choices

pgvector over a dedicated vector DB (Pinecone, Weaviate) — Supabase was already in the stack for relational data; adding the pgvector extension avoided a second service and kept the architecture simple without sacrificing search quality.

Server Actions over client-side fetch — requests to the backend happen server-to-server from Next.js, which sidesteps CORS entirely and keeps the API URL out of the browser's network tab.

Domain-specific RAG over generic document chat — rather than building another "chat with your PDF" clone, CareerLens solves one well-defined problem end to end, which made the demo concrete and the scope achievable.

---

Local development

Backend

cd backend
uv venv && source .venv/bin/activate
uv sync
cp .env.example .env   # fill in DATABASE_URL and GEMINI_API_KEY
uvicorn app.main:app --reload

Frontend

cd frontend
npm install
cp .env.example .env.local   # set NEXT_PUBLIC_API_URL=http://localhost:8000
npm run dev

Backend runs on localhost:8000, frontend on localhost:3000.

---

Running tests

cd backend
pytest -v

---

API reference

Method	Endpoint	Description
GET	/health	Health check
POST	/resume/	Upload resume text — chunks, embeds, stores it
POST	/analyze/	Submit a JD — returns structured match analysis

Full interactive docs (Swagger UI) auto-generated at /docs on the live API.

---

Project structure

careerlens/
├── backend/
│   ├── app/
│   │   ├── routers/        # health, resume, analyze
│   │   ├── services/       # embedding, retrieval, llm
│   │   ├── models/         # SQLAlchemy models (resume, chunk) + Pydantic schemas
│   │   ├── utils/          # text chunking
│   │   ├── main.py         # FastAPI app + router wiring
│   │   ├── config.py       # Pydantic settings
│   │   └── database.py     # SQLAlchemy engine + session
│   └── tests/              # pytest coverage on core pipeline
└── frontend/
    └── app/
        ├── components/     # ResumeForm, AnalyzeForm, ScoreRing, etc.
        ├── actions/        # Server Actions calling the API
        ├── lib/            # typed API client + shared types
        ├── utils/          # shared action helpers
        ├── layout.tsx      # root layout
        └── page.tsx        # main page

---

Notes on deployment

The backend runs on Render's free tier, which spins down after periods of inactivity — the first request after idle may take 30-60 seconds to respond while the service wakes up. Subsequent requests are fast.

The Supabase connection uses the transaction pooler (not direct connection) for the deployed backend, since Render's network doesn't support the IPv6 address Supabase's direct connection resolves to in some regions.

---

Built by

Ardhendu Pramanik — LinkedIn · GitHub · Portfolio

Built as a hands-on project to learn RAG, FastAPI, and Python backend development from scratch alongside an existing Next.js/React background.