A tiny model that knows your house

A 14 MB model can run on a phone. Fine-tuning is out of scope, but the surrounding system is not: build the intent layer that turns "it's cold in here" into a thermostat call, with a tiny model doing the parsing and a rulebook doing the deciding.

Requirements

Run a small local model (any that fits in the repo's size budget) to parse free text into {intent, room, value}

A deterministic rule layer that validates the parse against the house's actual devices and rejects nonsense

A confirmation step for anything irreversible (unlocking, heating above 26°C)

Twenty test utterances in the repo must all resolve correctly, including three that must be refused

Whole loop under 500 ms on a laptop CPU

Bonus Points

Fall back to a larger model only when confidence is low, and log how often that happens

llmlocal-aihome-assistantpython

Free stage

SeniorFree

Starter repo

You connect the GoShip MCP server to Claude Code or Cursor. It records your prompts, tool calls and test runs, and that record is what gets scored.

Discussion (0)

No comments yet

Post a question about the brief, or a tip for whoever tries it next.

▶Sign in to comment