These days LLMs are creating so much that writing manually feels nice, especially with all the typos and weird exporessions.
Recently I've been dealing with browser-based ML systems. There seems to be some amazing developments in the community specifically on HuggingFace regarding transformers.js and ONNX.
I tried making use of them in my projects. There are some hurdles. For example the requirements, WebGPU consistency and different browser versions allowing different things...
I believe in this approach as it is a BIG thing.
No backend-heavy, privacy by default, best client device wins!