How it Works — Two Tracks

How the head restaurant chef's flavor gets into the AI, and how that AI then operates at franchise locations — from both sides, in plain order.

1. Head-restaurant track — digitizing the flavor

The head restaurant chef wears a Galaxy XR and cooks as usual from a first-person perspective. Our AI learns from that footage to build a standard behavior model unique to that restaurant.

  • Invitation → partnership: Korean head restaurants reach us through a registered-mail QR or direct introduction, then enter the partnership process.
  • Systematized filming: 30–50 hours of head-restaurant footage, captured in phases with guidance per time-of-day, cooking step, and ingredient.
  • Automatic quality assessment: Each clip's quality score is automatically computed, managing the volume and quality of training data.
  • Model training: Multi-LoRA fine-tuning produces a model unique to that head restaurant.

2. Learner (franchisee) track — 1:1 comparison with the head restaurant

The franchisee chef wears Galaxy XR while cooking the same menu items, and the AI compares those motions with the head restaurant's standard in real time.

  • Theory phase: Before hands-on practice, voice-driven AI guidance covers the head-restaurant chef's know-how, ingredients, and tools.
  • Practice phase: Head-restaurant vs. learner motion compared at 0.1-second resolution — deviation scored across four dimensions (motion, time, ingredients, environment).
  • Periodic re-training: Learning is not one-and-done. Periodic re-training absorbs the small drifts that accumulate over time between the franchisee and the head restaurant's flavor.

3. What our AI is

"Our AI" is not a single model — it is an integrated system.

  • Gemini 2.5 Flash — multimodal LLM, fine-tuned on the head restaurant footage (training/inference core).
  • Gemini Nano — on-device AI running on Galaxy XR or paired smartphone (real-time evaluation, function calling).
  • YOLOX-S · OSNet x0.25 — object detection + tracker (NPU-accelerated).
  • Galaxy XR integration — passthrough camera + AR overlay deliver information into the learner's field of view.

The integrated design that binds these together into the cooking-transfer domain is itself the heart of our patent filing.

4. Device / form-factor roadmap

We distinguish honestly between what is operational today and what will be added later.

  • Current — Phase 1 — Galaxy XR full-stack: smartphone (platform hub) + Galaxy XR (real-time engine), with shared learner visuals and AR overlay as the Phase 1 target, in progress.
  • Phase 2–3 — smart-glasses form-factor expansion: Galaxy Glass jinju, micro-LED AR (Haean), and similar light-weight devices — added as optional form factors once they ship and pass our performance validation (currently roadmap).

For technology stack detail, see the Technology page. Security and privacy are covered on the Privacy page.