I'm trying to build HouseOS - a simulation platform where AI agents learn to manage an autonomous household. It started as a single Q-learning agent deciding when to use a battery or pull from the grid. The plan is to keep expanding it - one concept at a time - until it covers the full resource flows of a house.
The idea: whenever I learn something new, I build it. Knowledge becomes implementation. Implementation becomes understanding.
A Q-learning agent managing household battery storage. It decides when to use the battery, when to draw from the grid, and when to shed load - learning to minimize cost and carbon emissions.
Currently adding solar PV generation.
Energy is just the start. The plan is to keep layering in new systems - EVs, heat pumps, water, biogas, food production, carbon flows, hydrogen - until the simulation covers everything a household touches. Eventually: multiple agents managing subsystems cooperatively, physics-based models replacing simplified ones, and a digital twin that runs its own experiments.
Early and actively expanding. Personal learning project, so rough edges are expected.