Hello, I'm
24 years shipping production firmware and edge systems — from semiconductor fab automation in Japan to on-device AI inference at 46ms on Android. Specialised in the full path from C/C++ embedded to quantised neural models running with no cloud dependency.
Senior Software Engineer with 24+ years of experience across Embedded Systems, IoT, Industrial Automation, Semiconductor, and Medical domains. Specialized in C/C++ embedded development, device SDKs, and edge systems. Proven track record in building scalable IoT platforms and industrial systems across 5 countries.
Hands-on experience in Applied AI including multilingual NLP model training, deployment, and mobile integration — with published models on HuggingFace. Leverages AI-assisted engineering tools (Claude Code, Antigravity) to accelerate development and improve system quality. Currently at Pairpoint — the Economy of Things joint venture between Vodafone and Sumitomo.
AI-Powered Emotion Journal — Live on Google Play
Published Android app. Runs a fine-tuned RoBERTa model entirely on-device — INT8 quantised to 120MB, inferring 28 emotions at ~46ms on Snapdragon 8 Elite (no cloud). Investigated NPU delegation on Hexagon DSP via Qualcomm AI Hub; ONNX INT8 CPU outperforms LiteRT NPU for transformer workloads due to subgraph fragmentation. Full-database AES-256 encryption via SQLCipher. Multilingual: English (bundled), German/Japanese/Korean (downloadable). Stack: Kotlin, Jetpack Compose, ONNX Runtime, Android Keystore.
German Emotion Classification Model
Fine-tuned XLM-RoBERTa-base for 28-emotion detection in German text. Trained on translated GoEmotions + 10K native GermEval samples. F1 0.447, top-1 accuracy 69%.
Japanese Emotion Classification Model
Fine-tuned XLM-RoBERTa-base for 28-emotion detection in Japanese text. Trained on translated GoEmotions + 35K native WRIME samples. F1 0.396, top-1 accuracy 56%.
Korean Emotion Classification Model
Fine-tuned XLM-RoBERTa-base for 28-emotion detection in Korean text. Trained on translated GoEmotions + 50K native KOTE samples. F1 0.568, top-1 accuracy 68%.
“Tojo is a phenomenal human being, engineer, mentor, leader, and more. In any of these roles, I would not be surprised to find Tojo in the top few within any organisation. He always loved to experiment and innovate new things. He was an inspiration for the whole team and I would love to work with Tojo again.”
“I have worked very closely with Tojo during the last 16+ years. His technical contributions at multiple Japanese sites were of great value. His software design skills have been exemplary and he could quickly convert his ideas into proof of concepts. I was also particularly impressed by his team management skills and the way he always cultivated budding talents.”
“Tojo always has a sense of passion in learning new technologies and ensures that he creates forums to impart it to his team. He gives his team a platform for coming up with innovative ideas. He always tries to learn and unlearn — that is one of the best characteristics I look up to.”
Graduated with Distinction — a strong technical foundation that launched a 25+ year career in software engineering.