const me = {
location: "Kathmandu, Nepal 🇳🇵",
role: "Software Developer / Computer Vision",
currentFocus: "Real-time computer vision in the browser & full-stack apps",
funFact: "I taught the browser to read Devanagari ✍️",
askMeAbout: ["web dev", "computer vision", "database design", "Salesforce"],
};- 🔭 Currently working on Lekh Patro — a Nepali typing, OCR, translation & Bikram Sambat calendar PWA, now also a native Android app with home-screen widgets
- 🌱 Learning more ML / computer vision and on-device AI (Transformers.js, ONNX)
- 👯 Open to collaborating on open-source projects
- 🔒 Recently went back through a five-year-old project of mine and wrote up every bug I found — 20 of them
- 💬 Ask me about JavaScript/TypeScript, Python, computer vision in the browser, or why your booking system oversells
- ⚡ Fun fact: Salesforce Trailhead Double Star Ranger — 289 badges & 100k+ points 🏅
| Project | Description | Tech |
|---|---|---|
| ⌨️ Lekh Patro | Romanized Nepali typing IME with in-browser OCR, English↔Nepali translation (optionally fully on-device) & a Bikram Sambat calendar — also ships as a native Android app with home-screen widgets — live | React TypeScript Transformers.js Kotlin |
| 👁️ Human Anomaly Detection | My final-year project, restored and rebuilt — real-time fall/pose detection: YOLOv8 pose model plus a geometric posture classifier calibrated over 4,924 images, running in the browser or server-side; replaced a 3-class detector I trained on 8,340 images that evaluation showed keyed on scene appearance — live demo | YOLOv8 ONNX Node.js |
| ✋ Hand Gesture Recognition | Browser-only real-time recognition of 7 hand gestures — live | MediaPipe WASM React |
| 🎨 Air Canvas | Draw in the air with your finger — hand-tracked painting, no backend — live | MediaPipe Canvas API React |
| 🏟️ Stadium Booking | My 3rd-year project, restored — and fixed what I found: an admin panel open to anyone, and a race condition that oversold seats — live · the numbers | PHP 8 MySQL Docker |