I'm a computer science graduate with about 10 years of experience in Japan's automotive industry. My background is mainly embedded software for HMI systems, battery control software, CAN, and related vehicle communication technologies. I'm considering a move into AI and may eventually look for opportunities in Europe or Australia.
I'm unsure whether to focus first on AI theory and fundamentals, such as algorithms and how models work, or on practical skills like deploying models, optimizing them for devices, building AI applications, and working with agents and existing tools. I'm not aiming to become an ML researcher, but I also don't want to only call APIs without understanding the underlying technology.
Would Edge AI or another area that combines embedded systems and machine learning be a sensible transition path? I'd also appreciate advice on the skills and projects I should prioritize, the long-term career prospects, and how much my Japanese and fairly limited English might affect opportunities. I'm nearly 35, hold a highly skilled visa in Japan, and don't have children, so I'm relatively flexible about relocating.
3 Answers
It helps to separate two very different career directions. Developing new models is closer to applied mathematics, statistics, and research engineering. Building products with existing models, agents, and APIs is more like software engineering with AI components. Neither is automatically better paid; compensation depends heavily on your engineering depth, the industry, and your ability to deliver reliable systems.
For an application-focused route, learn enough fundamentals to make informed decisions: linear algebra and probability at a practical level, optimization basics, neural-network concepts, embeddings, retrieval, evaluation, and common failure modes. Then spend most of your time building complete projects with testing, monitoring, security, cost controls, and deployment. A portfolio that shows you can turn an idea into a dependable system will usually be more useful than a collection of small API experiments.
Your strongest positioning may be as an embedded or automotive engineer who can add machine learning to real products, rather than as a generic beginner in AI.
Your automotive and embedded background could give you a strong entry point into Edge AI. Knowledge of vehicle communication, embedded constraints, and safety-oriented systems is valuable, and it may be more realistic than competing directly with new graduates for general machine-learning roles. Japan may also have demand for people who understand both automotive systems and AI deployment.
For a starting path, learn the practical machine-learning workflow: Python, model training basics, evaluation, quantization, inference optimization, and deployment with tools such as ONNX Runtime or TensorRT. Build projects that run on constrained hardware rather than only making cloud demos. You still need enough theory to understand data quality, overfitting, latency, accuracy tradeoffs, and why a model behaves the way it does, but you probably do not need research-level mathematics at first.
International relocation is possible, though visa requirements and the geographic concentration of jobs can make it slower than expected. Your existing specialization may help more than trying to present yourself as a completely new AI engineer.
That makes sense. Edge AI does seem like a more natural transition given my embedded software and vehicle-communication experience. I’ll look more closely at deployment, optimization, and the tools used in production rather than focusing only on theory.
More information about your background matters, but based on what you’ve shared, you already have a useful foundation: a computer science degree, roughly a decade of embedded experience, HMI and battery-control work, and substantial CAN knowledge. That is a much stronger starting point than someone entering AI with no industry experience.
Before choosing a destination, compare specific job descriptions in Europe and Australia. Look for roles involving embedded machine learning, perception, robotics, intelligent vehicles, automotive data, or ML infrastructure. This will show you which technologies and language requirements appear repeatedly. English will matter for most international roles, so improving technical communication, documentation, and interview skills should be part of the transition. Japanese ability may limit some local roles, but it does not erase the value of your technical experience.
A practical plan would be to choose one target role, study its requirements, create two or three relevant projects, and then apply your existing automotive experience to those projects. Also check visa sponsorship and eligibility early, since relocation rules can be as important as technical qualifications.
I’m leaning toward the practical side and don’t plan to become an ML researcher. I’ll aim to learn enough theory to understand what I’m deploying, while putting most of my effort into complete projects and production concerns.