Job Post Highlights 🔍 🎯 📊

LLM-based Chrome extension that evaluates job postings against your resume using AI — extract key details, score relevance, and surface skill gaps in seconds using on-device and cloud LLMs.

Full Demo

Demo Stills

---

✨ Features

• Multi-Provider AI — Google Gemini (Cloud), Ollama (Local), or Chrome's On-Device Gemini Nano
• 2-Stage Pipeline — Stage 1 extracts structured job details; Stage 2 scores relevance against your resume
• Hybrid Extraction — Detects missing fields after Stage 1 and runs a targeted refinement pass for accuracy
• Pre-Warmed Model — On-device Gemini Nano session initialized at startup; subsequent evals use clone() for fast execution
• Privacy First — API keys and resume data stay in your browser (chrome.storage.local)
• Premium UI — Glassmorphism design, dark/light themes, side panel + pop-out window modes, timing breakdown
• Keyboard Shortcut — Cmd+J (Mac) / Ctrl+J (Windows) to open instantly
• Relevance Scoring — 0–5 scale with match/gap analysis, leveling notes, and unique insights

---

🚀 Quick Start

git clone https://github.com/hellosaumil/JobPostHighlightsExtension.git

1. Open chrome://extensions/ → Enable Developer mode
2. Click Load unpacked → Select the cloned folder
3. Pin the extension for quick access
4. Open Settings (⚙️) → Choose your AI provider and configure
5. Navigate to any job posting → Click Evaluate Relevance

See Setup Guide for detailed provider configuration.

---

🏗️ Architecture

The extension uses a 2-stage AI pipeline to minimize token usage and maximize accuracy:

flowchart LR
    A["🌐 Job Page"] --> B["📄 Extract DOM Text"]
    B --> C["⚡ Stage 1<br>Pre-Extract"]
    C --> D["🧠 Stage 2<br>Relevance Analysis"]
    D --> E["📊 JSON Output"]
    E --> F["🎨 UI Render"]

    subgraph Stage1["Pre-Extraction"]
        direction TB
        C0{"Pre-parse Settings"}
        C0 -->|"On-Device"| C1["Prompt API / Summarizer API<br>(Gemini Nano)"]
        C0 -->|"Model Provider"| C2["Cloud / Ollama<br>(Strict Schema)"]
        C1 -.->|"fallback"| C3["Regex Cleaner"]
        C2 -.->|"fallback"| C3
        C1 & C2 --->|"missing fields"| C4["Hybrid Refinement<br>(Main Provider + Offset)"]
    end

    subgraph Stage2["Relevance Scoring"]
        direction TB
        D1["On-Device"] ~~~ D2["Gemini Cloud"] ~~~ D3["Ollama"]
    end

    C --> C0
    Stage1 --> D
    D --> Stage2

    style Stage1 fill:#1a1a2e,stroke:#4a9eff,stroke-width:2px,color:#fff
    style Stage2 fill:#1a1a2e,stroke:#10b981,stroke-width:2px,color:#fff

Loading

See Architecture Guide for the full pipeline breakdown, scoring rubric, and output schema.

---

📁 Project Structure

JobPostHighlightsExtension/
├── manifest.json           # Chrome MV3 extension manifest
├── ai_service.js           # 2-stage AI pipeline, session mgmt, hybrid refinement
├── content.js              # DOM text extraction (content script)
├── background.js           # Service worker, side panel toggle, Ollama CORS bypass
│
├── prompts/
│   ├── stage_1.md          # Stage 1 extraction prompt (fields, rules, examples)
│   └── stage_2.md          # Stage 2 scoring rubric template
│
├── sidepanel.html/js       # Side panel UI + controller
├── window.html/js          # Pop-out window UI + controller (tab selector)
├── styles.css              # Shared styles (dark/light themes, glassmorphism)
│
├── assets/                 # Extension icons (16/48/128px + SVG)
└── docs/                   # Detailed documentation
    ├── setup.md            # Installation & provider config
    ├── ollama.md           # Ollama troubleshooting & CORS
    └── architecture.md     # Pipeline deep-dive & output schema

---

📚 Docs

Document	Description
Setup Guide	Installation, provider configuration, On-Device AI setup
Ollama Guide	CORS troubleshooting, environment variables, recommended models
Architecture	2-stage pipeline, scoring rubric, output JSON schema

---

🛠️ Technologies

• Platform: Chrome Extension (Manifest V3)
• AI Providers: Gemini Cloud API, Ollama (Local), Chrome Built-in AI (Gemini Nano)
• Frontend: Vanilla HTML / CSS / JavaScript (no frameworks)
• Fonts: Space Grotesk, JetBrains Mono, Silkscreen
• Design: Glassmorphism, CSS custom properties, responsive layouts

---

Acknowledgements

Thanks to and for co-building this extension.

---

📄 License

MIT License — see LICENSE for details.