Translation is the easiest of the three to build, and you may not need to build it at all. The pipeline is just microphone, then speech-to-text, then translate, then caption, and each step is a commodity cloud API (Gemini, Google Cloud, or Azure Speech) that already covers Japanese, Thai, and English. A working two-way version is a few days of wiring. The real work is the factory noise, not the AI, so plan for a good noise-cancelling mic, low latency, and a Thai-capable engine. And remember Gemini Live on the phone and the RayNeo on the lens already do this for free, so a custom build is only for on-glass captions on a glass that lacks them, or for offline use.