ALL EXPERIMENTS05 / 07 — An intention travels through a system

Synapse

Conversation connected to tools, media and action

What does an assistant become when conversation can reach the physical and digital world?

MY ROLE
Concept, interaction, architecture, orchestration, integrations and deployment
CONTEXT
Independent development and operation across dedicated hosts
STATUS
Active personal system
SYNAPSE / HOMELAB EXPLORERINTERACTIVE MODEL · 01
DRAG TO ROTATE · CLICK TO EXPLORESYSTEM INTERPRETATION

One intention. The right path.

Synapse connects interfaces, models and tools. Orchestration turns a request into a route through the system’s different capabilities.

AN ILLUSTRATIVE MODEL OF THE SYSTEM AND ITS CONNECTIONS.

From input to experience.

THE INTERACTION LOOP
  1. 01

    Express

    Start with a voice message, text, image or other supported media.

  2. 02

    Route

    The interaction reaches the appropriate model, tool or specialist.

  3. 03

    Act

    Services respond or an agent carries out an approved task.

  4. 04

    Return

    A response, result or referenced report comes back to the conversation.

01 / The interaction

One conversation, several senses

I conceived Synapse to turn local AI capabilities into a usable assistant. Voice and messaging connect to tools, media services and approved home-control routines through Telegram and an ESP32 voice endpoint.

Working interactions include photo follow-ups, exact transcription, audio understanding and short-video analysis. Lightweight media references let a conversation return to an image across turns without uploading it again.

02 / The infrastructure

Give each part a role

I designed the interaction and architecture, built the orchestration and integrations, and deployed the system across dedicated hosts. A stable gateway connects clients to the model cluster, while speech and multimodal processing run on dedicated nodes.

Synapse keeps its own tool loop for interaction and orchestration. Hermes handles specialist research and development work, with isolated workspaces and previews deployed to a separate Docker host.

03 / The action

Keep the work inspectable

An idea expressed in conversation can become a working application, built, tested and made available by the agent. Research missions can retain their objective, evidence and state so the work can be followed up, cancelled or resumed.

The system remains an active personal experiment. It clarifies ambiguous references, reports unavailable services and retains human approvals for consequential operations. Operational investigation has been tested on real infrastructure; broader autonomy is still being developed.

WHAT THIS PROJECT EXPLORES

The assistant is the connection between an intention, the right capability and a result you can inspect.

Interaction design / Local AI / Voice / Distributed systems
Back to selected work Explore the full archive