AI Job Search Assistant 🔍

A full-stack, RAG-powered career platform that helps job seekers find relevant openings, understand how their skills stack up, and check whether their resume will survive an ATS (Applicant Tracking System) scan — all through a natural-language chat interface.

🔗 Live demo: https://advanced-job-search.vercel.app (Note: backend runs on a free-tier host and may take ~30s to wake up on first request)

---

What it does

• Chat-based job search — ask "What jobs match my Python and ML skills?" and get a grounded answer plus matching job cards, generated via a RAG pipeline over live job postings.
• Resume parsing — upload a PDF resume; the app extracts 200+ technical skills across 7 categories, estimates years of experience, and pulls contact info.
• ATS scoring — get a 0–100 score across five dimensions (formatting, section completeness, action verbs, quantified achievements, keyword match against a job description) with specific, actionable suggestions.
• Live job board — semantic search over indexed postings, refreshed on demand from a free job-board API.
• Analytics dashboard — in-demand skills, job source breakdown, and search activity over time, computed from real Postgres data.

---

Architecture

┌──────────────────────────────┐
│   Next.js 14 + TypeScript     │   Vercel
│   Tailwind CSS v4              │
└───────────────┬────────────────┘
                 │ REST (JSON)
┌───────────────▼────────────────┐
│        FastAPI backend         │   Render (Docker)
│  ├─ /api/chat   (RAG)          │
│  ├─ /api/jobs   (search/scrape)│
│  ├─ /api/resume (parsing)      │
│  ├─ /api/ats    (scoring)      │
│  └─ /api/analytics             │
└──────┬──────────────┬──────────┘
       │              │
┌──────▼──────┐ ┌─────▼──────────┐
│  Qdrant      │ │  PostgreSQL    │
│  (vectors)   │ │  (Neon)        │
└──────────────┘ └────────────────┘
       │
┌──────▼──────────┐
│  Groq API        │
│  (Llama 3.3-70B) │
└──────────────────┘

---

Tech stack

Frontend

• Next.js 14 (App Router) + TypeScript
• Tailwind CSS v4
• Recharts (analytics charts)
• lucide-react (icons)

Backend

• FastAPI + Pydantic
• SQLAlchemy ORM
• Groq API (Llama 3.3-70B) for RAG responses
• Qdrant for vector search
• FastEmbed (BAAI/bge-small-en-v1.5) for embeddings — chosen for its small footprint to fit free-tier memory limits
• pdfplumber for resume PDF parsing
• BeautifulSoup for job description cleanup

Data & infrastructure

• PostgreSQL via Neon (serverless Postgres)
• Qdrant Cloud (vector database)
• Docker (backend containerization)
• Render (backend hosting)
• Vercel (frontend hosting)
• Arbeitnow public API (job postings source)

---

Project structure

.
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app, CORS, startup hooks
│   │   ├── config.py            # env-based settings
│   │   ├── db/
│   │   │   ├── database.py      # SQLAlchemy engine/session
│   │   │   └── models.py        # Job, Resume, ATSScore, SearchLog
│   │   ├── routers/
│   │   │   ├── jobs.py          # scrape, search, list, stats
│   │   │   ├── resume.py        # upload, parse
│   │   │   ├── ats.py           # ATS scoring
│   │   │   ├── chat.py          # RAG chat endpoint
│   │   │   └── analytics.py     # dashboard aggregates
│   │   ├── services/
│   │   │   ├── scraper.py       # Arbeitnow job scraper
│   │   │   ├── embedder.py      # resume parsing + skill ontology
│   │   │   ├── ats_scorer.py    # ATS scoring logic
│   │   │   ├── reranker.py      # cross-encoder reranking (optional)
│   │   │   └── rag.py           # RAG orchestration (Groq + Qdrant)
│   │   └── vectorstore/
│   │       └── qdrant_client.py # embeddings + vector search
│   ├── requirements.txt
│   └── Dockerfile
│
└── frontend/
    ├── src/
    │   ├── app/
    │   │   ├── page.tsx          # chat page
    │   │   ├── jobs/page.tsx     # job board
    │   │   ├── resume/page.tsx   # resume upload + ATS
    │   │   ├── analytics/page.tsx
    │   │   └── globals.css
    │   ├── components/
    │   │   ├── Navbar.tsx
    │   │   ├── ChatWindow.tsx
    │   │   ├── JobCard.tsx
    │   │   ├── ResumeUpload.tsx
    │   │   ├── ATSScoreCard.tsx
    │   │   └── AnalyticsDashboard.tsx
    │   └── lib/api.ts            # typed API client
    ├── tailwind.config.js
    └── postcss.config.mjs

---

API overview

Endpoint	Method	Description
/api/chat	POST	RAG chat — semantic search + LLM-generated answer
/api/jobs/scrape	POST	Scrape job postings and index into Qdrant + Postgres
/api/jobs/search	POST	Semantic/hybrid search over indexed jobs
/api/jobs/list	GET	Browse indexed jobs
/api/jobs/stats	GET	Index + database stats
/api/resume/upload	POST	Upload and parse a resume PDF
/api/resume/{id}	GET	Retrieve a parsed resume profile
/api/ats/score	POST	Run ATS scoring against an optional job description
/api/analytics	GET	Aggregate stats for the dashboard

Full interactive API docs available at <backend-url>/docs (Swagger UI).

---

Running locally

Prerequisites

• Python 3.11+
• Node.js 18+
• Free accounts: Groq, Qdrant Cloud, Neon

Backend

cd backend
python -m venv venv
source venv/bin/activate      # Windows: venv\Scripts\activate
pip install -r requirements.txt

# create .env with:
# DATABASE_URL=postgresql://...
# QDRANT_URL=https://...
# QDRANT_API_KEY=...
# GROQ_API_KEY=gsk_...
# FRONTEND_URL=http://localhost:3000

uvicorn app.main:app --reload

Backend runs at http://localhost:8000 — Swagger docs at /docs.

Frontend

cd frontend
npm install

# create .env.local with:
# NEXT_PUBLIC_API_URL=http://localhost:8000

npm run dev

Frontend runs at http://localhost:3000.

Seeding job data

With the backend running, call POST /api/jobs/scrape via /docs with a body like:

{
  "queries": ["python", "developer", "engineer"],
  "location": "",
  "max_per_query": 15
}

---

Deployment

• Frontend: deployed on Vercel, root directory frontend, env var NEXT_PUBLIC_API_URL pointing at the backend.
• Backend: deployed on Render as a Docker web service, root directory backend. Environment variables: DATABASE_URL, QDRANT_URL, QDRANT_API_KEY, GROQ_API_KEY, FRONTEND_URL.
• Vector DB: Qdrant Cloud free-tier cluster.
• Database: Neon serverless Postgres free tier.

Notes on free-tier constraints

• Render's free tier (512MB RAM) cannot run sentence-transformers/torch — embeddings use FastEmbed with an ONNX-based model instead.
• Hybrid (keyword + vector) search and cross-encoder reranking are implemented but disabled by default in production to stay within memory limits; pure semantic search is used instead.
• The backend spins down after ~15 minutes of inactivity on Render's free tier — first request after idle may take 30-60 seconds.

---

Key design decisions

• FastEmbed over sentence-transformers: drops the torch dependency entirely, fitting comfortably within 512MB RAM while keeping a 384-dimension embedding model (BAAI/bge-small-en-v1.5).
• Skill ontology-based parsing: rather than relying solely on an LLM for resume parsing, a curated dictionary of 200+ skills across 7 categories (languages, ML/AI, frameworks, data, backend, cloud/DevOps, tools) enables fast, deterministic, and free skill extraction.
• Weighted ATS scoring: the five scoring dimensions are combined with configurable weights, and keyword matching dynamically re-weights when a job description is provided versus when it isn't.
• Decoupled architecture: frontend and backend are independently deployable services communicating over a typed REST API, mirroring real-world microservice patterns.

---

Future improvements

• Re-enable hybrid search + reranking on a higher-memory tier
• Add authentication and per-user saved searches / resume history
• Scheduled job-index refresh via GitHub Actions
• LinkedIn job source integration
• Cover letter generation from resume + job description

---

License

MIT