Implement Autosuggestions and Autocompletions in the Keyboard

[angrezichatterbox]

Terms

• I have searched open and closed feature requests
• I agree to follow Scribe-Android's Code of Conduct

Description

Refactor the current autosuggestion and autocompletion system to replace the existing database based approach with a hybrid Trie + LLM architecture similar to the AOSP keyboard. The Trie will handle efficient prefix-based candidate generation, while the LLM layer will provide contextual reranking and next-word prediction to improve relevance and accuracy. This change aims to enhance suggestion quality, contextual awareness, and scalability while maintaining low-latency performance suitable for mobile environments.

Contribution

I would be working on this. Happy to hear suggestions and improvements to the current plan :)

[andrewtavis]

@angrezichatterbox, are we good to close this? Feel free to close if we are :)

One final check, we're still reverting to alphabetical autocompletions via #409 if we don't have the files added in from #610, right? And the plan is we'll switch over autosuggestion backups and maybe the primary autosuggestions to trained language models in the future, correct?

[andrewtavis]

@angrezichatterbox, quick ping on the above :) Feel free to close this if we're still reverting to the alphabetical autocomplete in the current implementation! 😊