Mechatronics Engineer who codes across the stack — from 1.58-bit LLM training kernels to industrial IoT gateways to 3D web apps. I'm happiest in software and ML, with hardware/embedded depth from the mechatronics side (electronics, PLCs, sensors, real-time control).
I learn by building hard things end to end. The projects below are self-driven: from-scratch deep learning, reinforcement learning, computer vision, and full-stack web — most built and debugged on a single 8GB consumer GPU.
| Project | What it is | Stack |
|---|---|---|
| bitnet-rwkv-lm | A 0.15B-param RWKV-7 language model with BitNet 1.58-bit ternary weights, trained from scratch on an 8GB GPU — QAT, 2-bit packing, RFT fine-tuning, local chat app | PyTorch, CUDA, quantization |
| spiking-ff-jepa | A backprop-free learning study: spiking (LIF) neurons + Hinton's Forward-Forward + JEPA + int4 QAT, with a full ablation notebook and an honest negative result | PyTorch, Norse |
| ai-fastener-inspection | Edge-deployable machine vision for automotive fastener inspection (Fastener-YOLO) with a full ablation study and reproducibility scripts | Python, YOLO, Jetson |
| kalman-lstm-ppo-trader | An uncertainty-aware RL trader: a custom Kalman-LSTM cell feeding a RecurrentPPO agent, with Neural-ODE & Fourier-Neural-Operator ablations | PyTorch, stable-baselines3 |
| circuit-idle (live demo) | An F1-themed idle game with a pure, fully-tested simulation engine and a Three.js dashboard | TypeScript, Svelte, Three.js |
Hands-on industrial automation and IIoT integration (proprietary work, not public):
- Fieldbus & protocols: PLC integration over RS485 / Modbus; reverse-engineered an Omron proprietary serial protocol by traffic sniffing.
- IIoT stack: MQTT telemetry into ThingsBoard dashboards; Advantech WISE-4060 I/O modules; deployment of MIMOS WISP (Wireless Industrial Sensor Platform).
- Tooling: built custom SDKs/adapters to bridge legacy industrial equipment to modern IoT.
Languages: Python · TypeScript/JavaScript · C/C++ · (embedded) C ML/DL: PyTorch · quantization-aware training · reinforcement learning · computer vision · SNNs Systems: CUDA-constrained training · data pipelines · model inference & packing Web: Svelte · Three.js/WebGL · Vite · CI/CD (GitHub Actions) Hardware/Embedded: PLCs · RS485/Modbus · MQTT · IIoT gateways · sensors & real-time control
- 📧 hafizradzi8901@gmail.com
- 💼 LinkedIn: coming soon
Mechatronics × software. Always shipping something.