SK-Connect — Local AI Job Card Generator

2026 SaskHack Hackathon Project

A Chrome extension that turns any webpage into a safety-first training job card using a local AI model (Ollama). Designed for electrical utility field technicians to document procedures, generate training aids, and record voice notes — entirely offline with no cloud dependencies.

---

How it Works

1. Browse to any procedure, safety document, or utility portal page in Chrome.
2. Click the SK-Connect extension icon.
3. Click "Create Job Card (Local)" — the page text is sent to your local Ollama server.
4. A structured job card is generated with steps, PPE, safety notes, and acceptance checks.
5. The Field Notes hub opens with the card saved for review, along with voice recording tools.

Everything stays on your machine. No data leaves your computer.

---

Prerequisites

Install these before running:

Tool	Download	Notes
Node.js (v18+)	https://nodejs.org	For the local job card server
Ollama	https://ollama.com	Local LLM runner
llama3:8b model	ollama pull llama3:8b	Run after installing Ollama
Chrome or Edge	—	Chromium-based browser required

---

Setup

1. Pull the AI model

ollama pull llama3:8b

Make sure Ollama is running (it starts automatically on most systems after install, at http://127.0.0.1:11434).

2. Start the job card server

Linux / macOS:

./start.sh

Windows:

Double-click start.bat

Or manually:

cd server
npm install
npm start

The server runs at http://127.0.0.1:8787. Keep this terminal open while using the extension.

3. Load the Chrome extension

1. Open Chrome and go to chrome://extensions
2. Enable Developer mode (top-right toggle)
3. Click "Load unpacked"
4. Select the chrome-extension/ folder from this project
5. The ⚡ SK-Connect icon will appear in your toolbar (pin it for easy access)

---

Usage

Generating a Job Card

1. Navigate to a webpage with procedure or safety information.
2. (Optional) Select specific text to use just that section instead of the full page.
3. Click the SK-Connect extension icon.
4. (Optional) Paste a YouTube video URL in the input field.
5. Click "Create Job Card (Local)".
6. The card is generated and the Field Notes hub opens automatically.

Field Notes Hub

The hub has four tabs:

• Power Line — Text notes + voice recording for power line observations
• Substation — Text notes + voice recording for substation documentation
• General — Incident log template + safety reminders
• Card History — All generated job cards (up to 20), with export and copy tools

Card History

Every generated card is saved to your Chrome extension storage. From the Card History tab you can:

• Expand any card to review the full content
• Copy a card's JSON to clipboard
• Copy a card into your current notes tab
• Export all cards as a single .json file

---

Project Structure

sk-connect/
├── chrome-extension/         # Chrome extension (load this folder)
│   ├── manifest.json
│   ├── popup.html / popup.js # Extension popup
│   ├── icons/                # Extension icons (PNG)
│   ├── make-icons.html       # Re-generate icons if needed
│   └── field-notes/          # Field Notes Hub (tab page)
│       ├── index.html
│       ├── app.js
│       └── styles.css
│
├── server/                   # Local Node.js server (must be running)
│   ├── server.js
│   └── package.json
│
├── grid-docs/                # Standalone version of the notes hub
│   └── index.html / app.js / styles.css
│
├── Dummy Scrape Target/      # Test page for validating extension scraping
│   └── dummy_scrape_target.html
│
├── start.sh                  # Linux/macOS startup script
└── start.bat                 # Windows startup script

---

Configuration

Change the AI model — Edit server/server.js line 9:

const MODEL = "llama3:8b"; // change to any installed Ollama model

Other capable models to try:

• mistral:7b — Fast and capable
• phi3:mini — Very fast, good for structured JSON
• llama3:70b — Higher quality, requires more RAM

Change the server port — Edit line 155 in server/server.js:

app.listen(8787, "127.0.0.1", () => { ... });

Also update host_permissions in chrome-extension/manifest.json to match.

---

Data Storage

Data	Where stored	How to clear
Generated job cards	Chrome extension storage (chrome.storage.local)	"Clear History" in Card History tab
Text notes	Browser localStorage	"Wipe All Local Notes" in General tab
Voice recordings	Downloaded to your disk as .webm files	Delete files manually

---

Troubleshooting

"Error: Failed to fetch" in the extension popup

• Make sure the server is running (./start.sh)
• Check that Ollama is running: open http://127.0.0.1:11434 in your browser

"Ollama error 404" in server console

• The model isn't downloaded yet. Run: ollama pull llama3:8b

Extension not appearing after loading

• Make sure you selected the chrome-extension/ folder (not the root of the project)
• Check chrome://extensions for error messages

Job card generated but content is generic/fallback

• The LLM may have returned malformed JSON. Try again on the same page.
• Try selecting specific relevant text before clicking "Create Job Card"

Voice recording not working

• Click "Allow" when Chrome asks for microphone permission
• Check browser microphone permissions at chrome://settings/content/microphone

---

Safety Notice

This tool is for documentation and training purposes only. It is non-operational — it does not provide switching instructions, repair procedures, or authorizations to perform electrical work.

• Treat all conductors as energized unless confirmed otherwise by authorized control
• Maintain required approach distances at all times
• Do not touch downed lines or nearby objects
• Follow your organization's lockout/tagout and switching authority rules
• In emergencies, contact your utility control center / dispatch

---

Tech Stack

• Extension: Chrome Manifest v3, vanilla JS, no build step
• Server: Node.js + Express, local Ollama API
• AI: Any Ollama-compatible model (default: llama3:8b)
• Storage: chrome.storage.local + browser localStorage
• Audio: WebM via MediaRecorder API

---

Team

Built at the 2026 SaskHack Hackathon by:

• Harini Krishnaraj
• Sam Egunjobi
• Grace Martinez
• Anna Wiens
• Talia Iwanchysko

Local AI, no cloud.