Sriman-Kunda-056/VidyaEdge

Offline multilingual AI tutor powered by Gemma and Ollama, with image questions, generated diagrams, progress tracking, and a mobile-first Gradio PWA. # VidyaEdge — Offline AI Tutor > Multilingual AI tutor for students in low-connectivity regions. > Powered by **Gemma 4** running entirely on-device via **Ollama**. > Built for the Gemma 4 Impact Challenge — May 2026. ## Evidence at a glance | Verified from `app.py` | Count | | --- | ---: | | Explicitly named languages | **9** | | Callable tutor tools | **3** | | SQLite tables | **3** | | Seed curriculum entries | **3** | | Student/teacher tabs | **2** | | Functions in the single-file app | **22** | > **Prototype status:** no automated test suite is included. Student questions, > answers, and progress can be stored in SQLite, so do not use identifiable > student information without an appropriate privacy and retention policy. ## Preview The preview is extracted from the matching 8-second local demonstration video and shows the real application UI rather than a generated mockup. --- ## What it does - Answers academic questions in **Tamil, Hindi, English, Telugu, Bengali, Malayalam, Kannada, Marathi, Gujarati** — and any other language Gemma 4 understands - Accepts **photos** of handwritten homework (when using `gemma4:e4b`, the multimodal variant) - Generates **step-by-step diagrams** for math and science, using Gemma 4's native JSON-structured output (not regex templates) - Streams responses token-by-token for instant feedback - Runs **100% offline** after one-time model download - Installable as a **PWA / Android APK** for native phone deployment --- ## Quick start ```bash git clone github.com cd VidyaEdge # 1. Install Ollama from ollama.com ollama serve # leave this running in one terminal # 2. Pull the model (in another terminal) ollama pull gemma4:e2b # ~4 GB, fast, fits on phones # OR ollama pull gemma4:e4b # ~7 GB, multimodal (vision) # 3. Install Python dependencies pip install -r requirements.txt # 4. Run python app.py # 5. Open in any browser # → local …