👓 SpecScribe AR

Spatial Architecture & Decision Intelligence for Snap Spectacles

Key Features • System Architecture • Prerequisites • Required Tokens & Secrets • Quickstart Guide • Testing & Verification • Hackathon Compliance

---

💡 The Problem & The SpecScribe Solution

Technical system design sessions live and die on physical whiteboards. Engineers debate trade-offs, sketch microservices, and mutate databases at high velocity. Traditional meeting recorders and AI transcription bots capture what was spoken, but discard three critical dimensions:

1. Visual State Evolution: How the whiteboard topology evolved across iterations.
2. Temporal Decision Lineage: Why a queue, cache, or replica was added or rejected at timestamp $T$, and which speaker advocated for that trade-off.
3. In-Flight Architectural Critique: Single points of failure, unauthenticated endpoints, or missing rate limiters are caught days later instead of live during whiteboarding when changes are trivial.

How SpecScribe AR Solves This

                      PHYSICAL WORLD                          SPATIAL COMPUTING (SPECTACLES AR)
               ┌──────────────────────────┐                    ┌───────────────────────────────┐
               │  Physical Whiteboard     │   Periodic 480p    │  Neon Waveguide Overlay       │
               │  Marker Diagram Strokes  │ ─────────────────> │  Auto-Routed Draw.io DAG      │
               └──────────────────────────┘   Camera Capture   │  (Transparent #000000 bg)     │
                            ▲                                  └───────────────────────────────┘
                            │                                                  ▲
                   Natural Discussion                               Live Push / Pull Critique
                            │                                                  ▼
               ┌──────────────────────────┐   SIK Pinch Voice  ┌───────────────────────────────┐
               │  Engineering Team        │ ─────────────────> │  "Raising-Hand" AI Teammate   │
               │  Verbal Trade-offs & ASR │    Component Spawn │  Proactive Security & Arch Orb│
               └──────────────────────────┘                    └───────────────────────────────┘

• Hands-Free Whiteboard-to-Draw.io Digital Twin: Periodically samples camera frames, filters static frames via perceptual hashing ($<5\text{ms}$), extracts graph deltas via Gemini 2.5 Flash Vision, recalculates a Sugiyama hierarchical DAG layout with drawio-mcp, and renders a transparent AR overlay via a pre-warmed headless Chromium daemon in $60\text{ms}\text{--}140\text{ms}$.
• SIK Pinch Component Palette: Four spatial PinchButton controls ([◇ Decision], [▭ Service], [🛢 Database], [○ Terminal]) anchored at $1.15\text{m}$ focal depth. Pinch any button and speak a title to inject nodes directly into the live graph.
• The "Raising-Hand" Ambient AI Teammate: An ambient 3D orb floating in the peripheral FOV that visually raises its hand with a concise floating pill badge ("🚩 Auth Token Leak Risk", "💡 Use Pub/Sub Queue") when it detects flaws, or provides on-demand research on pinch.
• Post-Meeting Temporal Decision Scrubber & RAG Hub: A Next.js 15 dashboard backed by Supabase pgvector that replays the exact diagram evolution alongside speaker turns, decision rationales, auto-generated Markdown technical specs, and RAG Q&A.

---

以上為 README 的開頭部分。其餘內容請閱讀原文。 檢視完整 README ↗