Update PRD: QWEN-TTS-0.6B, Gradio/HF Spaces, gr.Server, Stitch UI

- Replace ElevenLabs with local QWEN-TTS-0.6B for voice cloning and TTS
- Replace React Native + GCP with Gradio app on Hugging Face Spaces
- Add gr.Server for frontend customization
- Add Google Stitch-inspired UI design patterns

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

mission.md

@@ -0,0 +1,74 @@
+# Mission
+
+## Product Vision
+
+**ReadBookMom** is a voice-first interactive audiobook app for children ages 0–7. The app reads books aloud, lets children ask questions mid-story and hear answers grounded in the book, and lets parents record their voice so their child hears *them* reading — even when they aren't there.
+
+## Problem
+
+Young children learn by asking "why?" and "what happens next?" — but passive audiobooks can't answer back. Parents want to be the voice their child hears, but can't always be present. Existing children's audiobook apps offer neither real-time Q&A nor voice personalization.
+
+## Core Product Loop
+
+```
+Listen → Interrupt with a question → Hear an answer → Resume
+```
+
+The app also proactively asks comprehension questions at chapter breaks to encourage active engagement.
+
+## Target Users
+
+| Role | Age | Relationship to App |
+|---|---|---|
+| Child listener | 0–7 | Primary interactor (voice, large buttons) |
+| Parent / guardian | Adult | Account owner, voice donor, progress viewer |
+| Educator / caregiver | Adult | Secondary — monitors engagement |
+
+## Key Outcomes (12-week MVP)
+
+- ≥ 25 curated titles for ages 0–7
+- Avg. session length ≥ 12 minutes with ≥ 2 voice interactions per session
+- Q&A helpfulness ≥ 80% (parent survey)
+- Comprehension quiz participation ≥ 50% of sessions
+
+## Success Metrics
+
+| Metric | Target |
+|---|---|
+| End-to-end voice latency (PTT release → first TTS audio) | ≤ 3.0s median, ≤ 5.0s p95 (local model) |
+| ASR round-trip | < 540ms target |
+| App load (Gradio Space) | ≤ 5s |
+| Audio playback start | ≤ 2.0s |
+| Crash-free sessions | ≥ 99.5% |
+| Uptime | ≥ 99.5% (Hugging Face Spaces) |
+| Concurrent users (MVP) | 50 (scale to 500, constrained by GPU) |
+
+## MVP Scope
+
+**In scope:**
+- Browse and play curated English audiobooks (Home, My Library, Categories screens)
+- Voice Q&A: child asks, app answers from book content only
+- Comprehension prompts at chapter breaks (1–3 per 10 min)
+- Parent account + child profile (no personally identifying child data)
+- Progress tracking, favorites, basic offline caching (chapter-level)
+- Parent voice cloning via QWEN-TTS-0.6B (local model, no external API dependency)
+
+**Out of scope for MVP:**
+- Open-domain conversation (all answers must cite book content)
+- Social features, user-generated content
+- Multi-language catalog beyond English
+- In-app content purchases
+
+## Guardrails
+
+| Constraint | Requirement |
+|---|---|
+| Content grounding | 80%+ of Q&A answers must cite or paraphrase in-book passages; out-of-scope queries declined gracefully |
+| Privacy | COPPA-compliant: verifiable parental consent, no behavioral ads, data minimization, parental deletion dashboard |
+| Voice cloning | Biometric data encrypted at rest; local inference only (QWEN-TTS-0.6B); purpose-limited to in-app narration; not used to train general models; deletable on request |
+| Child safety | Content filter and prompt guardrails active; no sensitive topics; answers in Pre-A1 language level |
+| Security | TLS 1.2+, bcrypt/Argon2 passwords, parameterized queries, JWT with rotation, KMS-managed secrets |
+
+## Differentiator
+
+Real-time, child-safe, two-way voice Q&A tied to each specific book — plus parent voice cloning — in a single app designed for ages 0–7.

tech_stack.md

@@ -0,0 +1,166 @@
+# Tech Stack
+
+## Architecture Overview
+
+A Gradio web application deployed as a Hugging Face Space, with local QWEN-TTS-0.6B for voice cloning and TTS. UI design informed by Google Stitch for polished interface patterns.
+
+```
+Gradio App (gr.Server — customized UI)
+  │         Hugging Face Space
+  │
+  ├── Content Backend (Python)
+  │       │
+  │       ├── PostgreSQL (book metadata, users)
+  │       └── Pinecone (RAG embeddings)
+  │
+  └── Voice Service (local)
+          │
+          └── QWEN-TTS-0.6B (TTS + voice cloning)
+```
+
+---
+
+## 1. Front-End (Gradio + gr.Server)
+
+| Layer | Choice | Details |
+|---|---|---|
+| Framework | Gradio 5.x | Hugging Face-native, Python-first UI |
+| Customization | `gr.Server` | Custom routes, middleware, and frontend overrides |
+| Hosting | Hugging Face Spaces (GPU) | Free tier or upgraded Space for GPU inference |
+| UI Design | Google Stitch-inspired | Modern card layouts, smooth transitions, child-friendly large-button interface |
+| Local dev | `gradio app.py` | Hot-reload; same app runs locally and on HF Spaces |
+
+**Key UI components:**
+- Audio player with playback controls (book narration)
+- Voice recording interface (parent voice samples)
+- Large-button child interaction panel
+- Book library grid (card-based, Stitch-style)
+- Q&A chat interface with voice input/output
+
+---
+
+## 2. Voice Cloning & TTS Service (Local)
+
+| Layer | Choice | Details |
+|---|---|---|
+| Model | QWEN-TTS-0.6B | Local 0.6B parameter TTS model with voice cloning capability |
+| Inference | Transformers / local GPU | Runs on HF Space GPU or local CUDA device |
+| Voice cloning | QWEN-TTS-0.6B zero-shot | Few-second reference audio → cloned voice |
+| TTS | QWEN-TTS-0.6B | Text-to-speech for narration and Q&A answers |
+| ASR | Whisper (local) or browser Web Speech API | Speech-to-text for child voice input |
+| Audio format | WAV/MP3 | Generated on-the-fly or pre-cached |
+
+**Voice cloning constraints:**
+- Min reference audio: 5s, recommended: 15–30s
+- Local inference only — no audio data sent to external APIs
+- `child_safe_mode = True`, `content_filter_enabled = True`
+
+---
+
+## 3. Content Backend (Python)
+
+| Layer | Choice | Details |
+|---|---|---|
+| Framework | Gradio app (integrated) | Backend logic co-located with Gradio app |
+| Language | Python 3.12 | |
+| LLM | OpenAI GPT-4 (`gpt-4`) | Q&A answer composition |
+| Embeddings | OpenAI `text-embedding-3-large` | 3072-dim vectors |
+| Vector DB | Pinecone | RAG retrieval over book passage chunks |
+| RAG framework | LangChain | Chunking (`size=1000, overlap=200`), retrieval (`k=5`) |
+| Auth | Hugging Face OAuth or simple login | Gradio built-in auth support |
+| Logging | structlog (JSON) | |
+
+---
+
+## 4. Database (PostgreSQL 17)
+
+Four tables:
+
+| Table | PK | Foreign Keys | Purpose |
+|---|---|---|---|
+| `users` | `user_id` (UUID) | — | User profiles (HF OAuth or local) |
+| `booktext` | `book_id` | — | Book metadata + full text |
+| `voicesample` | `voice_id` | `user_id → users` | Parent voice recording (local file path or HF dataset) |
+| `audiobook_cloned` | `audiobook_cloned_id` | `book_id`, `user_id`, `voice_id` | User-specific cloned audio; `processing_status`: pending → complete |
+
+Naming convention: `snake_case`.
+
+---
+
+## 5. Infrastructure (Hugging Face Spaces)
+
+| Component | Service | Config |
+|---|---|---|
+| App hosting | Hugging Face Spaces (Gradio) | `app.py` + `requirements.txt` |
+| GPU inference | HF Space GPU (T4/A10G) | For QWEN-TTS-0.6B local inference |
+| Model hosting | Hugging Face Hub | QWEN-TTS-0.6B weights auto-downloaded |
+| File storage | HF Datasets / local persistent storage | Audio files, book files |
+| Secrets | HF Space Secrets | API keys (OpenAI, Pinecone) |
+| Database | External PostgreSQL (e.g., Supabase, Neon) | Lightweight managed DB |
+
+---
+
+## 6. CI/CD
+
+| Pipeline | File | Trigger |
+|---|---|---|
+| App deploy | HF Spaces auto-deploy | Push to `main` branch (linked repo) |
+| Tests | `.github/workflows/test.yaml` | Push to main |
+
+---
+
+## 7. Local Development
+
+```bash
+# Install dependencies
+pip install -r requirements.txt
+
+# Run locally (same app as HF Space)
+python app.py
+# or
+gradio app.py
+
+# Access at http://localhost:7860
+```
+
+---
+
+## 8. Key External APIs
+
+| API | Used For | Auth |
+|---|---|---|
+| OpenAI | GPT-4 (Q&A), `text-embedding-3-large` (RAG indexing) | `OPENAI_API_KEY` |
+| Pinecone | Vector similarity search over book passages | `PINECONE_API_KEY` |
+| Hugging Face Hub | Model weights (QWEN-TTS-0.6B), Spaces hosting | `HF_TOKEN` |
+
+**Removed external dependencies (now local):**
+- ~~ElevenLabs~~ → replaced by QWEN-TTS-0.6B (local)
+- ~~LiveKit~~ → replaced by Gradio audio components
+- ~~Google Cloud (GCS, Cloud Run, Cloud SQL)~~ → replaced by Hugging Face Spaces
+
+---
+
+## 9. UI Design (Google Stitch-Inspired)
+
+| Pattern | Application |
+|---|---|
+| Card-based layouts | Book library grid, chapter selection |
+| Smooth micro-animations | Page transitions, button feedback |
+| Large touch targets | Child-friendly interaction (≥ 48px tap areas) |
+| Consistent color system | Warm, playful palette for ages 0–7 |
+| Responsive design | Works on desktop (parents) and tablet (children) |
+
+`gr.Server` enables custom CSS/JS injection and route handling for Stitch-style polish beyond default Gradio theming.
+
+---
+
+## 10. Tooling
+
+| Tool | Purpose |
+|---|---|
+| `black` | Python code formatting |
+| `isort` | Import sorting |
+| `flake8` | Linting |
+| `mypy` | Type checking |
+| `pytest` + `pytest-asyncio` | Testing |
+| `admintool/gutenberg_downloader.py` | Bulk download public-domain books from Project Gutenberg |