class Harshal:
def __init__(self):
self.background = ["Physics", "Game Development", "Machine Learning"]
self.stack = ["Next.js", "Supabase", "TailwindCSS", "Python"]
self.focus = "research-grounded ML projects, not tutorial clones"
self.freelance = "Full-stack web apps on Fiverr"
def philosophy(self):
return "rigorous eval > flashy demo — no leakage, no shortcuts"- 🔭 Building research-grounded ML projects — benchmarked, evaluated honestly, and demoed live, not tutorial-style clones
- 🧠 Physics background → drawn to physics-informed ML and simulation-grounded RL
- 🕹️ Started in game development — still shows up in a love for interactive demos over static notebooks
- 💼 Freelance full-stack developer on Fiverr, shipping Next.js + Supabase apps
- 🧪 Obsessive about honest evaluation — if a model hits 98% accuracy, I go find out why before I trust it
| Project | Description | Stack |
|---|---|---|
| 🌌 Galaxy10 SSL | Self-supervised learning CNN for galaxy morphology classification on Galaxy10 DECaLS | PyTorch, SSL, CV |
| 🐕 Quadruped RL Locomotion | Physics-based quadruped locomotion via PPO, grounded in DeepMimic / DeepLoco, with a live teacher demo | PyBullet, Stable-Baselines3, PPO |
| 🗄️ Text-to-SQL Engine | NL→SQL system benchmarked on the Spider dataset | CodeT5, SQLite, Streamlit |
Swap
#for the actual repo links once this is pushed.
Turning physics, games, and models into things that actually work.