Alachi24/Kalabari-AI

a language model # Kalabari AI AI-powered translation and language detection for Nigerian languages. This is a **pnpm monorepo** with two apps that run as one system: ``` ┌────────────┐ same-origin ┌──────────────────┐ server→server ┌────────────────────┐ │ Browser │ ───────────────────▶ │ Next.js (BFF) │ ────────────────▶ │ Spring backend │ │ (React) │ /api/translate… │ route handlers │ /api/v1/… │ AI gateway │ └────────────┘ └──────────────────┘ └─────────┬──────────┘ │ │ │ auth + history │ chat/completions ▼ ▼ ┌────────────┐ ┌──────────────┐ │ Supabase │ │ Model server │ │ (auth, DB) │ │ (Gemma, TBD) │ └────────────┘ └──────────────┘ ``` ## Responsibilities - **`frontend/` — Next.js 16 / React 19.** The UI plus a thin **BFF (backend-for-frontend)** layer. Route handlers under `app/api/*` call the backend **server-to-server** (the backend URL stays server-only). Supabase is the single owner of **auth** and **translation history**. - **`backend/` — Spring Boot.** A **stateless AI inference gateway** exposing `POST /api/v1/translate`, `/detect`, and `/generate`. It forwards to an OpenAI-compatible model server and caches translations in Redis. Until the fine-tuned model is deployed it runs an in-memory **mock client** (`MODEL_SERVER_MOCK=true`). The frontend's old in-browser Transformers.js inference was removed — translation/detection now have a single implementation, in the backend. ## Prerequisites - Node.js 18+ and **pnpm 9+** (`corepack enable` then `corepack prepare pnpm@9 --activate`) - Java 21 + Maven (for the backend) - A Supabase project (auth + history) - Optional: Docker (for backend + Redis via `docker-compose.yml`) ## Setup ```bash # 1. Install JS dependencies (workspace ro …