martinwu500@gmail.com

Hi, my name is

Martin Wu.

I build things with code & hardware.

Software engineer building ML pipelines, full-stack apps, and embedded firmware. Co-founding a stealth-mode AI startup and engineering firmware at UBC Solar.

martin@dev ~
$ |

01. About Me

I'm an engineering student at UBC who builds across the stack — ML infrastructure, full-stack web applications, and low-level embedded firmware. I like working on problems that are technically hard and actually matter.

Right now I'm co-founding a stealth-mode AI startup focused on making large language models run fully locally on consumer hardware — targeting privacy-constrained industries where cloud AI isn't an option. I'm also building safety-critical embedded firmware at UBC Solar.

Here are some technologies I've been working with:

  • Python
  • C / C++
  • JavaScript / TypeScript
  • PyTorch / transformers
  • llama.cpp / LoRA / QLoRA
  • React / Flask
  • STM32 / SPI / CAN bus
  • Docker / InfluxDB / Grafana
Martin Wu

02. Experience

Dec 2025 — Present

Co-Founder & Software Engineer · Stealth-Mode AI Startup

Vancouver, BC

  • Built two-stage data pipeline for local large language model fine-tuning: 8B model filters domain PDFs for relevance; 70B model synthesizes and quality-judges Q&A training pairs
  • Developed LLM-as-a-Judge evaluation framework against a held-out golden anchor set, tracking factual accuracy, hallucination rate, and embedding drift with a 90% pass threshold
  • Designed closed-loop retraining pipeline: model stash on quality regression, automated LoRA/QLoRA fine-tuning trigger, and human escalation with cache rollback after 3 audit failures
  • Implemented VRAM-aware task scheduler with mutex locking to sequence inference, sanity checks, and training without GPU resource conflicts
  • Reduced 70B model VRAM requirements from 640GB (8x A100) to 48GB (2x RTX 4090) via Q4_K_M quantization, enabling consumer-hardware deployment
PythonPyTorchtransformersllama.cppLoRA/QLoRALLM-as-a-Judge
Sep 2025 — Present

Embedded Systems Engineer · UBC Solar

Vancouver, BC

  • Engineered PT1000 RTD temperature sensing system with MAX31865, closing a critical motor telemetry gap identified after a competition failure
  • Reverse-engineered Altium PCB schematics to integrate MAX31865 hardware into the real-time driver dashboard monitoring stack
  • Developed STM32 firmware in C implementing SPI communication and MAX31865 register configuration
  • Achieved <1°C temperature measurement accuracy through ADC-to-temperature conversion, improving from multi-degree variance
  • Integrated sensor data over CAN bus to dashboard, enabling safety monitoring independent of base station
  • Configured Docker/Tailscale/InfluxDB/Grafana pipeline in WSL2/Linux via Sunlink radio for live vehicle diagnostics
CSTM32WSL2/LinuxDockerSPICAN busGrafana

03. Projects

Other Noteworthy Projects

Portfolio Website

Responsive personal portfolio with dark/light theme toggle, scroll-triggered animations, and automated CI/CD deployment pipeline via GitHub Actions.

  • HTML/CSS
  • JavaScript
  • GitHub Actions

Math Quiz Generator

Quiz platform with adaptive difficulty scaling and weighted operation selection across 10 levels. MathJax rendering with GitHub Pages deployment.

  • React
  • Flask
  • MathJax

Haar Wavelet Compressor

Haar wavelet transform algorithm for lossy RGB image compression using NumPy matrix operations. Desktop GUI with real-time parameter adjustment.

  • Python
  • NumPy
  • CustomTkinter

04. What's Next?

Get In Touch

Whether you have an opportunity, a project in mind, or just want to connect, I’d love to hear from you. My inbox is always open.

OR
martinwu500@gmail.com
(236) 518-9477 Vancouver, BC