I build ML systems and the software around them. Based in Bengaluru, India.

Halfspace - Trying to answer "Who plays like this footballer, and why?"

• 397 Premier League players · 3 learned representations · consensus-based ranking
• An LLM explains the match 0 it never influences the ranking itself
• Uses one season of aggregate stats, capturing what a player produced rather than where or how they produced it
• Next: event-level + multi-season data, embeddings served through a real vector store, and retrieval caching to prevent redundant web scraping
  -> repo

Find Your Route 0 A civic transit co-pilot that reads a physical ticket and works out what the trip actually costs you.

• OCR pipeline extracts and normalizes station details from scanned tickets
• Gemma generates route guidance, which is then compared against alternative transport options on cost + carbon impact
• FastAPI + Redis for the backend and caching, with a local Ollama fallback when the hosted model isn't reachable
• Next: rate limiting + per-user quotas, cache-focused load testing, and moving OCR/generation onto a queue so slow requests don't block fast ones
  -> repo

Backend systems, GenAI, LLMs, NLP and representation learning. I’m especially interested in making retrieval actually return the right thing, and in understanding how ML systems work beyond just the model itself. I’m also getting more into the systems side - things like caching, rate limiting and infrastructure - through projects.

I’m also spending more time going back to things I’ve already built and making them better instead of constantly starting something new.

SWE / ML internships and part-time roles.

"Do it now. Later is just a nicer word for never."