Skip to content
Tillett Solutions
← Back to overview

Planned design direction

Smart home, with a personal AI assistant

A design for bringing everyday routines and conversation together, with feedback guiding what improves next.

Explore the design

Smart home

One assistant, working together

The design brings requests, conversation and everyday coordination into one Personal AI assistant. Scripted automation and a personal small language model work together within it.

Personal AI assistant

Scripted automation

Defined scripts handle routine actions and repeatable steps, with clear conditions for what each action does.

Personal small language model (SLM)

The model helps understand requests, reason through context, hold a conversation and coordinate the relevant routine.

The model interprets a request and connects it to a routine; the scripts provide the defined steps to carry it out.

From a request to a routine

Illustrative example: “Set things up for a quiet evening.” Here is how a request connects conversation with a routine.

  1. Understand the request

    Clarify what “quiet evening” means when the request is unclear, then identify the intended routine.

  2. Use defined steps

    Match the request to a script for an agreed lighting setting, using the relevant integrations and permissions.

  3. Review what happened

    Review a correction such as “keep the reading light brighter” when refining the routine.

Feedback gives improvement a direction

The learning loop connects outcomes with improvements. Review a correction before changing a routine or evaluating a model update.

Planned Personal AI learning cycle: Data → Personal model → Action → Feedback → Improve → Data Planned feedback loop within one Personal AI assistant. Data informs the personal model; scripted automation carries out defined actions, including physical home actions. Reviewed feedback informs improvements and returns to data. Chat Data Personal model Action Feedback Improve Planned Personal AI learning cycle: Data → Personal model → Action → Feedback → Improve → Data Planned feedback loop within one Personal AI assistant. Data informs the personal model; scripted automation carries out defined actions, including physical home actions. Reviewed feedback informs improvements and returns to data. Chat Data Personal model Action Feedback Improve
  1. Data

    Relevant requests, outcomes and reviewed examples provide context.

  2. Personal model

    Use that context to understand and coordinate requests.

  3. Action

    Carry out the selected routine through defined scripts, including physical home actions.

  4. Feedback

    Compare outcomes and corrections with the original intent.

  5. Improve

    Check a revised script or candidate model update before use. Reviewed examples feed the next cycle.

Data → Personal model → Action → Feedback → Improve → Data. Chat is an entry point to the same assistant.