I work at the intersection of machine learning, AI accelerators, and software systems. My background is in probabilistic deep learning and digital signal processing, with research across speech, images, text, and single-cell biology.
I use AI with custom harnesses for compiler development and high-performance scheduling, with a focus on MXFP shared group exponents and real-time, low-latency inference.
Alongside my 6G AI ASIC work, I build open-source tools for AI-assisted engineering: coding-agent integrations, developer workflows, and model inspection tools. I care about clear interfaces, measured performance, and understanding what actually works.
🇫🇮 Finland · 🇻🇳 Vietnam | LinkedIn · X
| Project | Focus |
|---|---|
| ASM | My personal fork of OpenAI Codex for local coding-agent workflows. |
| Codex subagents for Claude Code | OpenAI models through the Codex CLI as native Claude Code subagents. |
| System1 Thinking | Adaptive reasoning-effort selection for Claude Code. |
| AI Colab | Compact VS Code metrics and diagnostic context for AI-assisted development. |
| netron-rs | An early Rust toolkit for parsing, querying, and laying out neural-network model graphs. |
| codex-acp | A Codex plugin and Rust daemon for working with Agent Client Protocol (ACP) coding agents, including Cursor and Copilot. |
Tools I work with: Python · Rust · TypeScript · PyTorch · TensorFlow
- ODIN: probabilistic deep learning, variational autoencoders, and representation learning.
- SISUA: semi-supervised generative models for single-cell data.
- bigarray: memory-mapped NumPy arrays for scalable data access and multiprocessing.
- Covid-Cough-Detection: research on transformers and self-supervised learning for cough-audio classification.
Teaching and workshops
I have shared practical material on machine learning, speech processing, and Bayesian inference through University of Eastern Finland courses and workshops.
- UEF summer schools: 2016 · 2017
- Speech processing workshop (2017) · Sequence learning workshop (2018)
- Bayesian inference (2017): PyMC and Edward.
- Python for Computational Intelligence (2017)
- Deep learning course (2019): seven assignments and three mini-projects.
Side projects and life outside code
- Neural wedding card: neural style transfer for my wedding-card design.
- VeriQA-bot: an after-dinner experiment in answering questions from Veri's blog content.
- VSCode Autoflake: remove unused Python imports and variables.
- Coconerd: code counts for nerds.
- Brick racing: an Android take on the classic game.
Outside engineering: climbing, mountaineering books, and the history of science.
Personal projects and opinions, not an employer statement.




