diff --git a/README.md b/README.md index 9d80137..952bc32 100644 --- a/README.md +++ b/README.md @@ -1,177 +1,297 @@ # Taleka -image +Taleka hero banner -## Demo & Resources +> A mobile platform for learning and preserving endangered indigenous languages of Southeast Asia — starting with Semai (Malaysia). + +--- + +## What's New (Refinement Changelog) + +These features and improvements were added after the original BorNEO HackWknd 2026 submission on May 9th. + +### Story Pipeline + +- Elders can now **publish verified recordings as stories** directly from the Sound Archive tab +- Stories are sourced from the Supabase `recordings` table — scenes are built from `verified_transcription` split by paragraph, with `verified_translation_ms` as per-scene subtitles +- **AI-generated cover art and background images** via the `generate-story-cover` edge function (OpenRouter / GPT Image 2), using the recording's description and verified transcription as the image prompt +- New `cover_url`, `bg_url`, and `is_published` columns on the `recordings` table with RLS policies allowing uploaders to publish their own verified recordings + +### Multi-Language Support & Dynamic Backgrounds + +- Users can select any of 8 indigenous ASEAN languages as their learning language: Tandia · Semai · Mlabri · Chong · Arta · Arem · Kristang · Moken +- **Note:** language selection is UI-level for this hackathon — only **Semai** has a full implementation (dictionary, AI coaching, transcription, translation). The other 7 languages are placeholders ready for future corpus and model integration. +- The home screen hero image updates dynamically based on the user's chosen language — both learner and elder variants, one per country +- Country flag metadata added to each language option (`LEARNING_LANGUAGE_METADATA`) +- 20 background images bundled via Vite `import.meta.glob` + +### AI Coach Improvements + +- Claude Agent SDK added as provider (previously using Gemini) +- Conversation threads are **persisted across sessions** — the coach remembers prior turns +- Authentication is now required for all coach requests +- Improved reliability: coaching sessions recover gracefully from provider timeouts + +### Profile Onboarding + +- New guided onboarding flow for first-time users: choose role (Learner / Elder), set display name, select learning language, and optionally upload an avatar +- Language preference is stored in the `profiles` table (`indigenous_language` column) and immediately reflected in the hero background and app routing + +--- -- [YouTube Video Submission](https://youtu.be/bcL9RvYezy4): [https://youtu.be/bcL9RvYezy4](https://youtu.be/bcL9RvYezy4) -- [Figma Hi-Fi Design](https://www.figma.com/design/eUx2kJ3dssfOGgNwyiCoaz/HackWeeknd?node-id=0-1&t=2P25Uv7cXfUSoD7c-1): [https://www.figma.com/design/eUx2kJ3dssfOGgNwyiCoaz/HackWeeknd?node-id=0-1&t=2P25Uv7cXfUSoD7c-1](https://www.figma.com/design/eUx2kJ3dssfOGgNwyiCoaz/HackWeeknd?node-id=0-1&t=2P25Uv7cXfUSoD7c-1) -- [Final Report](Report_Taleka_HACKWKND.pdf): [Report_Taleka_HACKWKND.pdf](Report_Taleka_HACKWKND.pdf) +## Demo & Resources -Taleka is a mobile language learning and cultural preservation platform designed to help protect endangered indigenous languages. Built for the BorNEO HackWknd 2026 (Theme: The Role of AI in ASEAN Social Impact), the app currently focuses on Semai, an indigenous language spoken by the Semai people of Peninsular Malaysia. +| Resource | Link | +| ------------------ | -------------------------------------------------------------- | +| YouTube Submission | https://youtu.be/bcL9RvYezy4 | +| Figma Hi-Fi Design | https://www.figma.com/design/eUx2kJ3dssfOGgNwyiCoaz/HackWeeknd | +| Final Report | [Report_Taleka_HACKWKND.pdf](Report_Taleka_HACKWKND.pdf) | -The platform serves two user types: learners who want to learn Semai, and elders who preserve and share their language through voice recordings. By combining community voices, interactive learning tools, and AI-powered features, Taleka creates an engaging environment for language preservation and acquisition. +--- ## Why Semai? -Semai is an Aslian language (Austronesian family) spoken by approximately 30,000 people in Peninsular Malaysia. It is classified as endangered by UNESCO, as children are no longer acquiring it as their first language. The Semai language has limited digital presence and is largely excluded from commercial AI and NLP tools. Taleka aims to bridge this gap by providing a dedicated digital infrastructure for Semai language preservation and learning. +Semai is an Aslian language spoken by approximately 30,000 people in Peninsular Malaysia. UNESCO classifies it as endangered — children are no longer acquiring it as their first language. It has limited digital presence and is excluded from commercial AI and NLP tooling. Taleka provides dedicated digital infrastructure for Semai preservation and learning, with a roadmap to expand to other ASEAN indigenous languages. + +--- + +## Architecture + +![Architecture diagram](https://mermaid.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-IFN0b3JlXG4gICAgVUkgLS0-IExpYlxuICAgIExpYiAtLT4gQXV0aFxuICAgIExpYiAtLT4gREJcbiAgICBMaWIgLS0-IEZuQ29hY2hcbiAgICBMaWIgLS0-IEZuVHJhbnNsYXRlXG4gICAgTGliIC0tPiBGbkNvdmVyXG4gICAgTGliIC0tPiBBSUhlbHBlclxuICAgIEFJSGVscGVyIC0tPiBPbW5pQVNSXG4gICAgT21uaUFTUiAtLT4gU2NvcmVyXG4gICAgU2NvcmVyIC0tPiBMZXhpY29uXG4gICAgTGV4aWNvbiAtLT4gREIiLCJtZXJtYWlkIjp7InRoZW1lIjoiZGVmYXVsdCJ9fQ) + +### Data Flow: Elder Recording → Story + +![Data flow diagram](https://mermaid.ink/img/eyJjb2RlIjoic2VxdWVuY2VEaWFncmFtXG4gICAgcGFydGljaXBhbnQgRSBhcyBFbGRlclxuICAgIHBhcnRpY2lwYW50IEFwcCBhcyBNb2JpbGUgQXBwXG4gICAgcGFydGljaXBhbnQgSGVscGVyIGFzIGFpLWhlbHBlciBwcm94eVxuICAgIHBhcnRpY2lwYW50IEFTUiBhcyBPbW5pQVNSIChIRilcbiAgICBwYXJ0aWNpcGFudCBTQiBhcyBTdXBhYmFzZSBEQlxuICAgIHBhcnRpY2lwYW50IEZuIGFzIGdlbmVyYXRlLXN0b3J5LWNvdmVyXG4gICAgcGFydGljaXBhbnQgR1BUIGFzIEdQVCBJbWFnZSAyXG4gICAgcGFydGljaXBhbnQgTCBhcyBMZWFybmVyXG5cbiAgICBFLT4-QXBwOiBSZWNvcmRzIGF1ZGlvXG4gICAgQXBwLT4-SGVscGVyOiBQT1NUIC9haS90cmFuc2NyaWJlXG4gICAgSGVscGVyLT4-QVNSOiA0LWxhbmd1YWdlIHBhcmFsbGVsIHJlcXVlc3RzXG4gICAgQVNSLS0-PkhlbHBlcjogNCBjYW5kaWRhdGUgdHJhbnNjcmlwdGlvbnNcbiAgICBIZWxwZXItPj5IZWxwZXI6IFNjb3JlIGFuZCBtZXJnZSBjYW5kaWRhdGVzXG4gICAgSGVscGVyLS0-PkFwcDogQmVzdCB0cmFuc2NyaXB0aW9uIGRyYWZ0XG4gICAgQXBwLT4-U0I6IFNhdmUgZHJhZnQgdG8gcmVjb3JkaW5ncyB0YWJsZVxuICAgIEUtPj5BcHA6IFJldmlld3MgYW5kIHZlcmlmaWVzIHRyYW5zY3JpcHRpb25cbiAgICBBcHAtPj5TQjogU2V0IGlzX3ZlcmlmaWVkPXRydWUsIHZlcmlmaWVkX3RyYW5zY3JpcHRpb25cbiAgICBFLT4-QXBwOiBDbGlja3MgUHVibGlzaCBhcyBTdG9yeVxuICAgIEFwcC0-PkZuOiBpbnZva2UgZ2VuZXJhdGUtc3RvcnktY292ZXJcbiAgICBGbi0-PlNCOiBGZXRjaCBkZXNjcmlwdGlvbiArIHZlcmlmaWVkX3RyYW5zY3JpcHRpb25cbiAgICBGbi0-PkdQVDogR2VuZXJhdGUgY292ZXIgaW1hZ2VcbiAgICBGbi0-PkdQVDogR2VuZXJhdGUgYmFja2dyb3VuZCBpbWFnZVxuICAgIEZuLS0-PlNCOiBTYXZlIGNvdmVyX3VybCwgYmdfdXJsLCBpc19wdWJsaXNoZWQ9dHJ1ZVxuICAgIFNCLS0-Pkw6IFN0b3J5IHZpc2libGUgaW4gU3RvcnlQYWdlIiwibWVybWFpZCI6eyJ0aGVtZSI6ImRlZmF1bHQifX0) + +--- ## Features ### Elder Studio -A voice recording and transcription tool that enables community elders to archive their spoken language. The interface is designed for elderly users with minimal technical experience, featuring one-tap recording and simple controls. Recordings are processed through speech-to-text (Whisper) to create searchable, learnable content for the community. +Voice recording and transcription for community elders. One-tap recording with waveform visualization. Recordings are transcribed using a multi-language OmniASR ensemble, then dictionary-scored and corrected. Elders review transcriptions and verify them against the Semai lexicon, which continuously expands the dictionary. -Elders can review and verify their own transcriptions against existing dictionary entries. New or alternative meanings are flagged and merged into the lexicon after a brief self-verification, continuously expanding and improving the dictionary. +### Publish as Story + +Verified recordings can be published as illustrated stories. The app generates a 2:3 portrait cover image and a full-screen background image using the recording's text as the visual prompt. Published stories are immediately visible to all learners. ### Language Garden -Interactive vocabulary learning with spaced repetition. Users build their Semai word bank through curated word lists organized by category, with definitions in English and Malay. The vocabulary is sourced from a verified Semai dictionary, enabling learners to cross-check vocabulary and sentences with authoritative meanings. +Spaced-repetition vocabulary learning. Words sourced from the verified Semai dictionary, organized by category. Progress tracked in Supabase (`progress` table, `mastery_level > 0` = learned). Leveling system (XP = wordsLearned × 4): Seed (0 XP) → Sprout (500 XP) → Sapling (1,200 XP) → Flourish (2,500 XP) → Legacy (4,000 XP). ### Stories -A collection of folk tales and stories in Semai, contributed by community elders. Learners can read stories with translations, track reading progress, and revisit completed stories. Stories are organized with metadata including elder contributor, theme tags, and reading duration. +Folk tales and stories contributed by community elders, now dynamically fetched from Supabase. Each story is broken into scenes by paragraph. Scene images, cover art, and backgrounds are AI-generated. Reading progress tracked locally and synced. -### AI Coach +### AI Coach (Tavi) -A personal AI tutor powered by a ChatGPT/Codex proxy with Gemini fallback that provides conversational practice in Semai. The coach uses the project's glossary and sentence examples to give grounded, contextually accurate responses. It supports multiple interaction modes including vocabulary help, sentence construction, translation assistance, and free conversation. +Conversational Semai tutor. Uses a deterministic rule layer (greetings, translation requests, known words) before hitting the LLM, so verified Semai is never invented by the model. Providers: Claude Agent SDK (primary) → Gemini (fallback). Sessions persist across app restarts. ### Translation -Bidirectional translation between Semai, English, and Malay. The translation system uses the project's dictionary and sentence memory for accurate, context-aware results. Translation requests are handled through Supabase Edge Functions. +Bidirectional Semai ↔ Malay ↔ English. Priority chain: exact glossary match → exact sentence example → LLM with glossary constraints → word-by-word fallback. The LLM is the last resort, not the first. ### Learning Games -- **VocabMaster**: Swipe-based vocabulary practice with card animations. Users swipe right to mark words as known, left to review again. Each round presents 10 cards with definitions in English and Malay. -- **WordleGame**: Daily Semai word guessing game where players have 6 attempts to guess a random word from the vocabulary. -- **QuizGame**: Interactive quizzes to test language comprehension across multiple categories. +- **VocabMaster** — swipe-based flashcard practice (10 cards/round, right = known, left = review) +- **WordleGame** — daily 6-attempt Semai word guess +- **QuizGame** — multi-category comprehension quiz ### Sound Archive -A searchable library of all recorded Semai audio. Users can browse recordings by category, search for specific words, and play audio with variable playback speed for learning purposes. +Searchable library of all recorded Semai audio. Browse by category, search by keyword, play at variable speed. ### Profile & Progress -User profiles track learning progress, display current level, earned achievements, and weekly leaderboard rankings. The leveling system rewards consistent engagement with the app. Users can customize their profile with avatar, display name, and language preferences. +XP = `wordsLearned × 4`. Weekly leaderboard, avatar upload, role-based navigation (Learners get 5 tabs, Elders get 4 (Home · Story · Record · Profile)). + +--- + +## Tech Stack -## Technical Stack +| Layer | Technology | +| ----------- | -------------------------------------------------- | +| Frontend | React 18, Ionic 8, Capacitor 6, Vite, TailwindCSS | +| State | Zustand | +| Backend | Supabase (Postgres, Auth, Storage, Edge Functions) | +| STT | OmniASR via Hugging Face (4-language ensemble) | +| AI Coach | Claude Agent SDK (primary), Gemini (fallback) | +| Translation | Cerebras (primary), SEA-LION (optional) | +| Image Gen | OpenRouter / GPT Image 2 (story covers) | +| Testing | Vitest + Testing Library | +| CI | GitHub Actions (lint, typecheck, test) | -image +--- ## Project Structure ``` . -├── LICENSE # MIT License -├── README.md # This file -├── ai-helper/ # OmniASR proxy for local transcription -├── docs/ # Planning and assessment documents -├── supabase/ # Database migrations and edge functions +├── ai-helper/ # Local OmniASR transcription proxy (Node.js, port 8787) +│ ├── server.js # HTTP server — STT ensemble, scoring, lexicon cache +│ └── semaiLexicon.js # Curated Semai term list +├── assets/ +│ ├── flags/ # Country flag SVGs (one per ASEAN indigenous language) +│ └── landing/background/ # Hero images — {country}-{learner|elder}.png (20 total) +├── supabase/ +│ ├── migrations/ # Postgres schema migrations +│ └── functions/ +│ ├── ai-coach/ # Conversational Semai tutor +│ ├── ai-translate/ # Semai ↔ Malay ↔ English translation +│ ├── generate-story-cover/ # AI cover + background image generation +│ ├── seed-words/ # Initial dictionary seed +│ └── _shared/ # Bundled Semai corpus (glossary + sentence examples) └── src/ - ├── components/ui/ # Reusable UI components - ├── lib/ # Core utilities (auth, supabase, storage, AI) - ├── pages/ # Page components - ├── stores/ # Zustand state stores - ├── types/ # TypeScript type definitions - └── utils/ # Validation and helper utilities + ├── components/ui/ # Reusable UI (AppButton, AppSkeleton, …) + ├── lib/ # Core logic (auth, supabase, AI, storage, learningLanguages, …) + ├── navigation/ # AppRouter — auth gates, role-based tab routing + ├── pages/ # All page components + per-page CSS + ├── stores/ # Zustand (authStore) + ├── types/ # TypeScript types (Supabase DB schema) + └── utils/ # Validation, auth errors, profile progress ``` -## Getting Started +--- + +## Setup ### Prerequisites - Node.js 22+ -- npm or pnpm +- A Supabase project (free tier works) +- (Optional) Hugging Face account for OmniASR transcription -### Installation +### 1. Clone & install ```bash -# Install dependencies +git clone https://github.com/pablonification/HACKWKND-2026.git +cd HACKWKND-2026 npm install - -# Start development server -npm run dev ``` -### Running AI Helper (Local Transcription) +### 2. Configure environment -Elder Studio transcription uses `VITE_AI_BASE_URL` and calls `POST /ai/transcribe`. +Copy the example and fill in your values: -Run the proxy locally: +```bash +cp .env.example .env +``` + +| Variable | Description | +| --------------------------------- | ---------------------------------------------------------- | +| `VITE_SUPABASE_URL` | Your Supabase project URL | +| `VITE_SUPABASE_ANON_KEY` | Supabase anon/public key | +| `SUPABASE_SERVICE_ROLE_KEY` | Service role key (used by ai-helper proxy) | +| `VITE_PASSWORD_RESET_REDIRECT_TO` | Redirect URL after password reset | +| `VITE_AI_BASE_URL` | Transcription proxy URL (default: `http://localhost:8787`) | + +### 3. Run database migrations ```bash -npm run ai-helper:dev +npx supabase db push ``` -Then set frontend env: +### 4. Deploy edge functions ```bash -VITE_AI_BASE_URL=http://localhost:8787 +npx supabase functions deploy ai-coach +npx supabase functions deploy ai-translate +npx supabase functions deploy generate-story-cover +npx supabase functions deploy seed-words ``` -### Building for Mobile +Set required secrets: ```bash -# iOS -npm run cap:ios +# AI Coach +supabase secrets set GOOGLE_AI_STUDIO_API_KEY=your_key +supabase secrets set AI_COACH_GEMINI_MODEL=gemini-3.1-flash-lite-preview -# Android -npm run cap:android +# Translation (Cerebras) +supabase secrets set CEREBRAS_API_KEY=your_key + +# Story cover generation (OpenRouter) +supabase secrets set OPENROUTER_API_KEY=your_key +``` + +### 5. Start the frontend + +```bash +npm run dev +# → http://localhost:5173 ``` -### Running Tests +### 6. Start the transcription proxy (Elder Studio) + +Required for voice recording and transcription in Elder Studio: ```bash -npm test # Watch mode -npm run test:run # Single run +npm run ai-helper:dev +# → http://localhost:8787 ``` -### Code Quality +### 7. Build for mobile ```bash -npm run check # Full check (lint, format, typecheck, test) +# iOS (requires macOS + Xcode) +npm run cap:ios + +# Android +npm run cap:android ``` -## AI Configuration +--- -### AI Coach (Supabase Edge Function) +## AI Coach — Advanced Configuration -The personal AI coach uses the `ai-coach` edge function and reuses the project glossary and sentence-memory assets for grounded Semai responses. For the competition demo, it can use a protected local Claude Agent SDK gateway as the primary provider with Gemini as fallback. +The `ai-coach` edge function supports a Claude Agent SDK gateway as the primary provider for demos: -Set these edge-function secrets for coach generation: +```bash +# Edge function secrets +supabase secrets set AI_COACH_PROVIDER_ORDER=claude-agent,gemini +supabase secrets set AI_COACH_CLAUDE_AGENT_BASE_URL=https://your-gateway.example.com/v1 +supabase secrets set AI_COACH_CLAUDE_AGENT_API_KEY=your_key +supabase secrets set AI_COACH_CLAUDE_AGENT_MODEL=sonnet +``` + +Run the local gateway (requires a Claude Code–authenticated machine): ```bash -AI_COACH_PROVIDER_ORDER=claude-agent,gemini -AI_COACH_CLAUDE_AGENT_BASE_URL=https://your-claude-agent-gateway.example.com/v1 -AI_COACH_CLAUDE_AGENT_API_KEY=your_taleka_claude_agent_gateway_key -AI_COACH_CLAUDE_AGENT_MODEL=sonnet -GOOGLE_AI_STUDIO_API_KEY=your_google_ai_studio_key -AI_COACH_GEMINI_MODEL=gemini-3.1-flash-lite-preview +TALEKA_CLAUDE_AGENT_KEY=your_key npm run demo:claude-agent-gateway ``` -Optional overrides: +The gateway exposes `/v1/models` and `/v1/chat/completions` and must be tunnelled to a public HTTPS URL (e.g. ngrok) for Supabase to reach it. + +Optional tuning knobs: ```bash -AI_COACH_DIRECT_MAX_OUTPUT_TOKENS=220 -AI_COACH_PEDAGOGY_MAX_OUTPUT_TOKENS=150 -AI_COACH_INTENT_MAX_OUTPUT_TOKENS=96 -GOOGLE_AI_STUDIO_BASE_URL=https://generativelanguage.googleapis.com/v1beta -AI_COACH_TIMEOUT_MS=12000 -AI_COACH_PROVIDER_RETRY_COUNT=1 -AI_COACH_PROVIDER_RETRY_DELAY_MS=900 +supabase secrets set AI_COACH_DIRECT_MAX_OUTPUT_TOKENS=220 +supabase secrets set AI_COACH_PEDAGOGY_MAX_OUTPUT_TOKENS=150 +supabase secrets set AI_COACH_TIMEOUT_MS=12000 +supabase secrets set AI_COACH_PROVIDER_RETRY_COUNT=1 ``` -Local Claude Agent SDK gateway: +--- + +## Development ```bash -TALEKA_CLAUDE_AGENT_KEY=your_taleka_claude_agent_gateway_key \ -TALEKA_CLAUDE_AGENT_RETRY_COUNT=2 \ -TALEKA_CLAUDE_AGENT_RETRY_DELAY_MS=900 \ -npm run demo:claude-agent-gateway +npm run dev # Vite dev server (port 5173) +npm run ai-helper:dev # Transcription proxy (port 8787) +npm run typecheck # TypeScript check (no emit) +npm run lint # ESLint (zero warnings) +npm run lint:fix # ESLint with auto-fix +npm run format # Prettier +npm run test # Vitest watch mode +npm run test:run # Vitest single run +npm run check # Full quality gate (lint + format + typecheck + test) +npm run precommit # Pre-commit gate (runs automatically via Husky) ``` -The gateway forwards only `/v1/models` and `/v1/chat/completions` into the Claude Agent SDK. It must run on a machine authenticated with Claude Code or configured with supported Claude credentials, then be exposed to Supabase through a protected HTTPS URL for demos. +All changes to `main` go through a PR. Greptile AI reviews every PR; minimum confidence score: 4/5. + +--- ## License -MIT License — see [LICENSE](LICENSE) for details. +MIT — see [LICENSE](LICENSE) for details. diff --git a/assets/landing/background-elder.png b/assets/landing/background-elder.png deleted file mode 100644 index 403a2d5..0000000 Binary files a/assets/landing/background-elder.png and /dev/null differ diff --git a/assets/landing/background-learner.png b/assets/landing/background-learner.png deleted file mode 100644 index 562b8e0..0000000 Binary files a/assets/landing/background-learner.png and /dev/null differ diff --git a/assets/landing/background/brunei-elder.png b/assets/landing/background/brunei-elder.png new file mode 100644 index 0000000..d85b875 Binary files /dev/null and b/assets/landing/background/brunei-elder.png differ diff --git a/assets/landing/background/brunei-learner.png b/assets/landing/background/brunei-learner.png new file mode 100644 index 0000000..e0f8a3b Binary files /dev/null and b/assets/landing/background/brunei-learner.png differ diff --git a/assets/landing/background/laos-elder.png b/assets/landing/background/laos-elder.png new file mode 100644 index 0000000..122b214 Binary files /dev/null and b/assets/landing/background/laos-elder.png differ diff --git a/assets/landing/background/laos-learner.png b/assets/landing/background/laos-learner.png new file mode 100644 index 0000000..0880968 Binary files /dev/null and b/assets/landing/background/laos-learner.png differ