Upgrade default model to MiniCPM-V 4.6; compact output panel into tabs

Deploy student-first exam coach

Remove read-aloud (VoxCPM2) feature entirely; drop voice deps (app.py + requirements.txt)

Fix voice: main-process VLM eviction before VoxCPM2 + dedicated voice panel

Fix: CSS-driven wizard step display (reliable hide under gradio 6)

Fix: observer-based JS at load; pure server updates for wizard events

UX: less text, genuinely interactive wizard (auto-advance, smart nav, JS scrolling/animations)

Perf: resident MiniCPM-V + boot warm-up + live status + honest fallback reasons (app.py & study_engine.py together)

UI: interactive guided rescue flow (Next/Back step wizard) — app.py only

UI: guided Rescue flow step card + coach next-step hint (app.py only)

UI: privacy-first note + prominent MiniCPM-V-4.5 photo upload (app.py only)

Llama Champion: selectable llama.cpp engine (small MiniCPM4 0.5B GGUF via llama-cpp-python, CPU)

MiniCPM-V-4.5 is now the primary engine (text + photo); drop 4.1-8B and 0.5B/1B; Nemotron-4B alternate; README to match

Add Field Notes build-report link to the footer

One text model at a time (24GB safety) + voice gating + pinned model metadata

Upload app.py with huggingface_hub

Resume Space

Prefetch model weights on CPU before the GPU call so a cold secondary model no longer times out the ZeroGPU budget

Remove claim-status panel (hard-tested on staging, 8/8)

UI cleanup: force light mode + declutter + fix duplicate IDs (validated on staging)

Harden GPU endpoints (graceful fallback on node faults)

Add drill answer keys (validated on staging)

Add VoxCPM2 read-aloud (5th model; validated on staging)

Add download/print packet button (validated on staging)

Add model picker: MiniCPM-8B / NVIDIA Nemotron-4B / MiniCPM5-1B (validated on staging)

Add MiniCPM-V syllabus-snap (vision) + torchvision (validated on staging)

Live coach timer: ship coach_state + UI together (validated on staging)

Revert to stable app.py (coach wiring caused a runtime error)

Add live coach mode (interactive triage clock)

Fix: raise Minutes-left slider max to 1440 so the full-day sample works

Isolate GPU call (graceful fallback on worker failure) + skip ZeroGPU preload

Expose generate as a stable API endpoint for real-model trace generation

Sync from GitHub: triage time-budget fix, MiniCPM-driven drills, UI polish, sub-4B model option, LICENSE

[Codex] Fix sample loaders in Space

[Codex] Gate sample generation for final Space

Improve Space text contrast with Codex

Record verified ZeroGPU MiniCPM smoke with Codex

Make Space product-first and ZeroGPU-ready with Codex

Focus MiniCPM demo and readable UI with Codex

Polish UI and hardware honesty with Codex

Deprioritize Cohere claim from Codex 3cdf800

Simplify demo-first Space UI

Deploy Exam Panic Rescue staging Space