I'm a third-year ECE student at IIIT Bangalore, working on FPGA architecture and hardware accelerators for machine learning.
I like problems where nobody's mapped the gaps yet. Most of what I build starts from reading a paper, noticing something that doesn't sit right, and chasing it until it turns into an idea worth testing.
- INSITU: a fixed-function attention accelerator on FPGA that scores and accumulates directly over a compressed KV cache, no reconstruction needed
- INSTINCT: a reinforcement learning policy for zero-shot eFPGA floorplanning, replacing per-design search with a single rollout
- 8-bit Breadboard CPU: a working CPU built from discrete TTL logic and coordinated Arduinos, no pre-made cores
- Currently reading into processing-in-memory and memory-centric architecture
- Peeling back abstraction layers until you hit something fundamental
- Circuits and systems that actually do something, not just simulate
- The moment an idea clicks after weeks of not seeing it
- Guitar, when the code isn't cooperating
building things, and figuring out what's actually worth building next