• ABOUT
• ARCHITECTURE
• ROADMAP
• DEMO

ABOUT

Dakota is an experimental AI assistant powered by Dakota AI Response Layer or DARL 0.1 - a custom micro-language model trained on a small, focused dataset of human-like greetings and social prompts. While still in its early phase, Dakota showcases how micro-models can deliver fast and lightweight responses. Our current approach proves the model works offline or locally. With the right investment, we’ll bring that to the cloud for speed and reach. We will eventually explore both online and hybrid deployments.Note: This project is currently in its early prototype phase, built entirely by a solo developer using Flutter for the mobile interface and FastAPI (Python) as the backend server.

ARCHITECTURE

Flutter for the cross-platform mobile interface of Dakota AI (currently Android only)FastAPI (Python) for handling local inference and API communication.NanoGPT as the training framework for DARL, our custom micro-language model.

ROADMAP

The project's goal is to explore and experiment with lightweight, small-scale AI model before moving on to something bigger. This early version allows for testing how micro-models like DARL 0.1 perform in real-world conversational tasks. The long-term vision is to release DARL 1.0 - a larger and more capable model once the core framework and data layer have matured.- DARL 0.1 (Current)
Trained on ~50-200 greeting and conversational prompts. Focused on tone, friendliness and fast response.- Coming Soon (0.2+)
-> Expanded prompt categories: daily habits, tasks and social replies.
-> Better tone matching (based on the mood/emotion)
-> API integration plans
-> Optional whisper/speech input- Goals for DARL 1.0
-> Multi-turn conversation handling
-> Modular response engine with plug-and-play prompt packs
-> Basic personalization

CHAT WITH DAKOTA - DEMO

Will be uploaded soon.