Projects & Repositories

Full-stack hardware-software co-design framework for Edge FPGAs — custom NPU architecture, proprietary ISA, and CUDA-like driver/API to accelerate Transformers and neural networks.

Lightweight LLM serving stack focused on low-latency inference and compact deployment.

LLM inference accelerator on AMD Kria KV260 — custom NPU IP on FPGA with Vitis HLS.

Real-time driver drowsiness detection using computer vision and deep learning.