I build and study systems at the intersection of efficient AI, digital hardware, and developer tooling.
My current work focuses on reproducible LLM evaluation under edge-hardware constraints, FPGA/ASIC design, embedded systems, and production-minded software projects.
| Project | What it does | Stack |
|---|---|---|
| qquantqqwen | Reproducible study of quantization, sparsity, and behavioral robustness in Qwen models on edge hardware | Python, PyTorch, llama.cpp |
| Goshorty | URL shortener with TTL, operational documentation, testing, containers, and CI | Go, Docker, GitHub Actions |
| ttsky-7seg-counter | Tiny Tapeout BCD counter with cocotb verification and an OpenLane GDS flow | Verilog, cocotb, OpenLane |
| cubsol | Full-stack application with typed data access, tests, CI, and Vercel deployment | TypeScript, Next.js, Drizzle |
| espqr | ESP32 QR-code generator for OLED and ST7789 TFT displays | C++, Arduino |
| Floyd_Warshall | Interactive all-pairs shortest-path visualization with path reconstruction | JavaScript, Cytoscape.js |
- Efficient and trustworthy LLM inference on constrained hardware
- Quantization, pruning, behavioral evaluation, and reproducible research
- Verilog/SystemVerilog, FPGA verification, and open-source silicon flows
- ESP32, embedded C/C++, and edge AI
- Agent-friendly CLI tooling and reliable automation
Python · PyTorch · Go · TypeScript · C/C++ · Verilog/SystemVerilog
GitHub Actions · Docker · Linux/WSL · Vivado · llama.cpp · Ollama
- Prefer reproducible results over one-off demos.
- Keep experiments, release artifacts, and large model files clearly separated.
- Test claims with paired comparisons, explicit manifests, and automated checks.
- Treat documentation as part of the implementation.
📍 Ho Chi Minh City, Vietnam