Electronics engineer working on AI hardware acceleration and reinforcement learning for optimizing and controlling RF, wireless, and autonomous physical systems. First-author research published on embedded fall detection.
Currently part-time Assistant Lecturer teaching Computer Architecture, Computer Programming, and Computer Graphics, and conducting VHDL-based digital design labs.
My work spans the full stack, PCB design and firmware up through model training and on-device deployment.
- AI hardware acceleration: computer architecture, synthesizable hardware accelerators
- RL for wireless systems: reinforcement learning for spectrum access and antenna optimization (see
dqn-dynamic-spectrum-access,rl-dipole-tuning) - Wearable health-IoT: embedded ML for real-time sensor data (fall detection, activity recognition)
PhD opportunities and research collaboration in AI accelerator architectures, computer architecture, reinforcement learning for intelligent wireless and autonomous physical systems, and edge AI.
Khatri, S. K., Parajuli, D., Mane, P. M., Chaulagain, B., & Thapa, S. (2026). Fall Detection System for Elderly People Using LSTM. Journal of Engineering Issues and Solutions, 5(1), 210-220.
Bachelor in Electronics, Communication & Information Engineering, Pulchowk Campus, Institute of Engineering, Tribhuvan University (First Division with Distinction). Registered Engineer, Nepal Engineering Council.
| Project | What it is |
|---|---|
scaled-dot-product-attention |
Verilog accelerator for scaled dot-product attention using two shared MAC units, Q8.8 fixed-point datapath, and LUT-based softmax. |
rl-dipole-tuning |
RL environment wrapping NEC2 for automated dipole antenna resonant-frequency optimization |
dqn-dynamic-spectrum-access |
PyTorch replication of Wang et al. (2018) DQN for dynamic multichannel access; benchmarked against the analytical optimal policy and a Whittle Index heuristic on a correlated-channel POMDP |
Fall-Detection-System |
Wearable fall-detection device + double-layer LSTM model (97.8% accuracy) - collaborative project; companion code to the publication above; designed the wearable PCB, collected the training dataset |
Startracker-Simulator-for-Attitude-Determination-of-Spacecrafts |
Synthetic star-image simulator and visualization layer for spacecraft attitude determination - collaborative project; built the image generation & visualization layer |
Languages: Python C/C++ VHDL Verilog
Embedded: STM32 ESP32 RP2040
ML/DL: PyTorch TensorFlow
Hardware: KiCAD PyNEC
Cloud/Dev: AWS Git Linux