ResumeLens: An AI Resume Screener & Job Matcher

A production-grade NLP web app that scores how well a resume matches a job description, surfaces missing skills, and recommends improvements.

Architecture: FastAPI backend + custom HTML/CSS/JS frontend (no framework bloat). Uses sentence-transformers for semantic embeddings and cosine similarity for matching.

---

Features

• Custom-designed UI — recruiter-friendly corporate aesthetic, light & dark mode
• Drag-and-drop PDF upload with size & filename preview
• Semantic similarity via all-MiniLM-L6-v2 (384-dim embeddings)
• Animated score gauge with dynamic color (red → amber → blue → green)
• Keyword gap analysis — see what's missing vs. matched
• Actionable suggestions based on score and gaps
• Fast — ~1–2s analysis after model warmup
• Local & private — your resume never leaves your machine

---

Project Structure

resumelens/
├── backend/
│   ├── main.py             # FastAPI app + REST endpoints
│   ├── nlp.py              # PDF extraction, embeddings, scoring
│   └── requirements.txt    # Python dependencies
└── frontend/
    ├── index.html          # Single-page UI
    ├── styles.css          # Design system (light + dark)
    └── app.js              # File handling, API calls, animations

---

Setup

1. Open PowerShell in the project folder

cd "C:\path\to\resumelens"

2. Create and activate a virtual environment

python -m venv venv
venv\Scripts\activate

3. Install dependencies

pip install -r backend\requirements.txt

(First install takes 5–10 minutes — PyTorch is large.)

4. Run the server

cd backend
uvicorn main:app --reload --host 0.0.0.0 --port 8000

5. Open the app

Visit http://localhost:8000 in your browser.

The first analysis will download the all-MiniLM-L6-v2 model (~80 MB) — one-time only.

---

Architecture

┌──────────────────────────────────────────┐
│    Browser (HTML / CSS / vanilla JS)     │
│  • Drag-and-drop upload                  │
│  • Animated score gauge                  │
│  • Light/dark theme toggle               │
└────────────────┬─────────────────────────┘
                 │  POST /api/analyze
                 │  multipart: resume.pdf + job_description
                 ▼
┌──────────────────────────────────────────┐
│         FastAPI (uvicorn)                │
│  • /api/analyze   – scoring endpoint     │
│  • /              – serves index.html    │
│  • /static/*      – CSS, JS              │
└────────────────┬─────────────────────────┘
                 │
                 ▼
┌──────────────────────────────────────────┐
│              NLP pipeline                │
│  PyPDF2 → text                           │
│  sentence-transformers → embeddings      │
│  scikit-learn → cosine similarity        │
│  Custom tokenizer → keyword gap          │
└──────────────────────────────────────────┘

Why this split? The frontend is fully decoupled from the backend — you can deploy them together (single uvicorn process, as configured) or split them later (e.g., frontend on Vercel, API on Render).

---

API

POST /api/analyze

Request (multipart/form-data):

• resume: PDF file
• job_description: text string

Response (JSON):

{
  "score": 78.4,
  "label": "Strong Match",
  "missing_keywords": ["docker", "kubernetes", "fastapi"],
  "matched_keywords": ["python", "machine", "learning"],
  "suggestions": ["…", "…"],
  "resume_preview": "First 600 chars of extracted text…"
}

GET /health

Returns {"status": "ok"} — useful for uptime monitors.

GET /docs

Interactive Swagger UI auto-generated by FastAPI.

---

Design Notes

• Typography: Fraunces (display serif, with optical-size axis) + Inter Tight (body) + JetBrains Mono (technical labels)
• Palette: Warm off-white in light mode, deep navy in dark mode, with azure-700 as the single accent
• No gradients-as-decoration — gradients reserved for the brand mark and hero atmosphere
• CSS variables for the entire color system — adding new themes is trivial
• No frontend framework — keeps the bundle effectively zero, loads instantly

---

Possible Extensions

• LLM-based resume rewriting (OpenAI / Claude API)
• Multi-job batch comparison
• ATS simulation scoring
• Dockerfile for one-command deployment
• User accounts + history (PostgreSQL)

---

License

MIT — free to use, modify, and distribute.

---

Built by Muhammad Shaheer Akhtar · 2026