Orbit.

It is also the first project where I integrated local AI deeply, running the model on my own infrastructure instead of depending on a third-party external API. That let me build my own endpoints and a rule system that guides the model so its answers and decisions stay coherent with the user's real context. Working with local AI also brings a key privacy advantage: the user's financial data, health data, and personal notes never leave my own infrastructure to be processed by third parties — especially relevant when dealing with such sensitive information.

One of the parts I enjoyed most was connecting Orbit to Obsidian: I implemented a RAG-like system (retrieval-augmented generation) that indexes my notes and uses them as context, so the app's assistant can answer questions and give advice based on real, personal information — not just the model's generic knowledge.

Over time I expanded Orbit beyond finances and tasks: it now includes a health and goal-tracking module, with the intention of bringing in data from wearable devices for a fuller picture of physical activity and wellbeing. The end goal is for the AI to help manage time, health, and money more efficiently and personally, always prioritising the user's control over their own data.

Orbit is, ultimately, a personal lab where I test applied-AI ideas before taking them to other projects, and at the same time a tool I use myself every day. It is in production and in active use.