What's Next?
Future improvements and explorations. Some more likely than others. No promises, no timelines.
Exploring RAFT
Retrieval-Augmented Fine-Tuning. Combining the benefits of RAG with fine-tuning for more accurate, context-aware responses. The idea: train the model to better leverage retrieved context.
iMessage Group Support
Support for iMessage groups. Mainly on the search side but there's lots of things that I'd want to chat and embed them with, it's just understanding how.
Improving Fine-Tuning Quality
Better conversation parsing, smarter training data curation, hyperparameter optimization. The goal is more authentic persona generation with less training data and fewer artifacts.
App Store Distribution
Submit to the Mac App Store for easier distribution and updates. Requires sandbox compliance, notarization, and... explaining to Apple why this exists.
Multi-Modal Support
Incorporate image attachments from conversations. Train on memes, screenshots, and photos to understand visual context. Because a picture is worth a thousand tokens.
Voice Synthesis Integration
Generate voice responses using local TTS models. Hear your AI persona speak. Absolutely terrifying implications that I'm aware of.
Export & Portability
Export trained personas for backup or transfer. Standard model formats for interoperability. Your data, your models, your choice.