Nemotron Resume Scanner

AI-powered resume scanner for the space and aerospace industry. Upload a PDF resume and get an ATS compatibility score, missing keyword analysis, and an AI-rewritten PDF — all powered by NVIDIA Nemotron vision model.

Live version → (part of Zero G Talent, a job board for 50+ space companies)

---

How It Works

PDF Upload → pdfjs-dist → Page Images → Nemotron Vision → Structured JSON

Architecture

1. PDF to Images — Client-side rendering with pdfjs-dist. Each page is converted to a JPEG at 1.5x scale. Up to 3 pages analyzed.

2. Vision Analysis (Temperature 0.3) — Resume images are sent to nvidia/nemotron-nano-12b-v2-vl:free via OpenRouter. The model performs OCR and analysis in a single pass, returning structured JSON with:

   • ATS compatibility score (0-100)
   • Formatting quality score
   • Keyword density score
   • Space industry fit score
   • Section-by-section breakdown
   • Specific improvement suggestions
   • Missing keywords for aerospace roles

3. Resume Rewriting (Temperature 0.4) — A second Nemotron call takes the original images + analysis feedback and generates an improved resume as structured JSON, which is rendered to a PDF using jspdf.

4. Job Matching — Keywords extracted from the analysis are matched against 30 sample aerospace jobs using simple text matching. In the live version, this searches across 10,000+ real job listings.

Why Nemotron?

nvidia/nemotron-nano-12b-v2-vl is a 12B-parameter vision-language model optimized for document understanding. It can:

• Read text from complex resume layouts (multi-column, tables, graphics)
• Understand document structure (headers, sections, bullet points)
• Extract and evaluate content in a single pass
• Return structured JSON consistently at low temperature

The model is available free on OpenRouter, making it ideal for open-source projects.

Two-Call Pattern

The scanner uses two separate API calls with different temperatures:

Call	Temperature	Purpose
Analysis	0.3	Consistent, structured scoring — needs to be deterministic
Rewriting	0.4	Creative rewording — needs slight variation for natural language

This split avoids the trade-off between consistency (scoring) and creativity (rewriting).

---

Quick Start

# Clone
git clone https://github.com/mishafyi/nemotron-resume-scanner.git
cd nemotron-resume-scanner

# Set up environment
cp .env.example .env
# Edit .env and add your OpenRouter API key (free at https://openrouter.ai)

# Install and run
npm install
npm run dev

Open http://localhost:3000 and upload a resume PDF.

Get an API Key

1. Go to openrouter.ai and sign up (free)
2. Create an API key
3. Add it to .env:

   OPENROUTER_API_KEY=sk-or-v1-your-key-here
   

The Nemotron model is free on OpenRouter — no credit card required.

---

Project Structure

src/
├── app/
│   ├── layout.tsx              # Root layout (dark theme, Space Grotesk font)
│   ├── page.tsx                # Main scanner page
│   ├── globals.css             # Tailwind + custom styles
│   └── api/
│       ├── analyze/route.ts    # POST: resume → analysis JSON
│       ├── improve/route.ts    # POST: resume + analysis → rewritten resume
│       └── match-jobs/route.ts # POST: analysis → matched sample jobs
├── components/
│   ├── ResumeUploader.tsx      # Drag-and-drop PDF uploader with pdfjs
│   ├── AnalysisResults.tsx     # Score rings, breakdowns, PDF generation
│   └── ui/                    # shadcn/ui components (badge, button, separator)
├── lib/
│   ├── resumeAnalysis.ts       # Nemotron prompt engineering + response parsing
│   ├── sampleJobs.ts           # Simple keyword search over sample data
│   └── utils.ts                # cn() utility
└── data/
    └── sample-jobs.json        # 30 sample aerospace jobs

Key Files

• src/lib/resumeAnalysis.ts — The core of the project. Contains the system prompt, vision message builder, and response parser. This is where the Nemotron prompt engineering lives.

• src/app/api/analyze/route.ts — Server-side API route that calls OpenRouter. Handles rate limiting, input validation, and error handling.

• src/components/ResumeUploader.tsx — Client-side PDF processing. Uses pdfjs-dist to render PDF pages to canvas, then converts to JPEG data URLs.

---

Tech Stack

Layer	Technology
Framework	Next.js 15 (App Router)
Frontend	React 19, Tailwind CSS 4, TypeScript
AI Model	NVIDIA Nemotron Nano 12B V2 VL (via OpenRouter)
PDF Processing	pdfjs-dist (client-side)
PDF Generation	jsPDF (client-side)
UI Components	shadcn/ui (badge, button, separator)

No database. No authentication. No payments. Just clone, add a key, and run.

---

NVIDIA GTC / DGX Spark

This project was built for the NVIDIA GTC DGX Spark contest. It demonstrates:

• Nemotron vision model for document understanding
• Structured output from a vision model (JSON parsing)
• Two-temperature pattern for analysis vs. creative generation
• Real-world application — resume analysis for the space industry

---

License

MIT — see LICENSE.

Built by Misha as part of Zero G Talent.