Dev Project: The Neural Glide: LLM-Driven Android Input Method Strategy

My phones keyboard typing is often terrible for mistyping, I set about developing a better one is one of my projects, I have a work in progress like 80% built for it. Building your own Android keyboard is a classic "weekend project" that can quickly turn into a "month-long obsession." Adding an LLM for context-aware suggestions makes it significantly more complex, but using modern AI tools actually makes the coding part much more approachable. This project proposal outlines the creation of the Android Smart Keyboard, a custom input application that replaces standard dictionary predictions with a Large Language Model. By leveraging a hybrid architecture, the keyboard primarily runs a quantized local model on the device’s Neural Processing Unit to ensure user privacy and offline functionality. To manage hardware constraints on the target device, the system only triggers AI inference during specific pauses or actions, while a cloud fallback remains available for complex tasks. Developers plan to fork existing open-source code from the Android Open Source Project to handle complex typing geometry, integrating the AI to select the most contextually relevant words. The development process will be further accelerated by using AI coding agents to manage complex boilerplate code and debugging. Overall, the initiative seeks to modernise mobile typing by making glide-typing significantly more intuitive through sentence-aware logic. I chose the Honor 200 LTE as simply it was a cheap reasonable pwoerful entry level mobile I have access to test such a thing on, it may well be that a more higher tier premium model might be a better candidate for it, which I will likely find out in testing, if I ever get to that stage.