Eye
The Eye Loop solves taste.
Core trick
Swipe accept/reject decisions on AI-generated probes are mined for emergent taste axes, which a builder agent uses to assemble a live HTML prototype.
Description
The Eye Loop solves taste. The user types an intent. Swipes through a real-time curated mood board. And gets a prototype built from their taste. It fully leverages Gemini's multi-modal capabilities to bridge the gap between human vision and AI understanding.
README excerpt
# The Eye Loop
Swipe to discover your taste. AI builds what you actually want.
The Eye Loop is a multi-agent taste discovery system. You type an intent ("finance app that doesn't feel like a spreadsheet"), swipe through AI-generated probes, and watch as the system discovers your aesthetic preferences and builds a prototype that reflects your actual taste — not just what you said.
## How it works
1. **Enter an intent** — describe what you want to build
2. **Swipe** — accept or reject visual probes (words, images, mockups)
3. **Watch your taste form** — the Anima panel shows emergent taste axes discovered from your choices
4. **See the prototype evolve** — the builder agent assembles a live HTML draft from your revealed preferences
## Architecture
Three agent types run concurrently on the server:
- **Scouts** (Iris, Prism, Lumen, Aura, Facet, Echo) — 6 taste probes running concurrently using the Akinator pattern. Each has a sensory lens (visual, layout, narrative, mood, info design, motion). Iris and Aura pre-buffer images during the word stage so moodboards are ready when you need them.
- **Builder** (Meridian) — assembles an evolving HTML prototype from accepted/rejected evidence. Freezes the draft at reveal. Emits construction-grounded probe briefs to guide scouts.
- **Oracle** — runs a cold-start intent analysis on session init, then discovers emergent taste axes from evidence every 4 swipes. Assigns each scout to a different axis. Detects persona-anima divergence (w