tinotendamaisiri/neriah-ai-public

Neriah AI — Gemma 4 powered homework grading + on-device tutor for African classrooms. Kaggle Gemma 4 Good Hackathon submission. # Neriah AI Neriah AI is an AI-powered homework marking platform for African classrooms. Teachers provide homework questions; Gemma 4 generates the marking scheme automatically. Students submit handwritten work through the mobile app, WhatsApp, or email. Gemma 4 marks each submission against that scheme and builds a weakness profile per student from the results. That weakness profile then drives two adaptive learning tools: a Socratic tutor that guides without giving answers, and Neriah Play, an arcade-mode study game that generates 100 questions from the student's own curriculum and dynamically adjusts game speed based on performance (+5% on correct, -5% on wrong). Teachers get a dashboard to review AI grades, approve results, and query their class data in natural language. Marking creates the data; the data powers the adaptation. Built for the Gemma 4 Good Hackathon on Kaggle (deadline May 18, 2026). ## Try the app | Platform | Link | |---|---| | Android APK | drive.google.com | | iOS (TestFlight) | testflight.apple.com | | Web | neriah.ai | Demo accounts (no signup needed): Teacher: +1 347 445 9587 — OTP: 000000 Student: +1 123 456 7890 — OTP: 000000 ## Architecture ## Screenshots ## What's in this submission This repo IS the submission. The hackathon entry is the working product, not a walkthrough notebook: a Cloud Functions backend (`functions/` + `shared/`), a React Native + Expo mobile app (`app/mobile/`), the dataset-extraction pipelines that power the on-device fine-tune (`tools/dataset/`), and a pytest suite that locks the contracts (`tests/`). Build and run instructions are in the `Running locally` section below. ### Core technology - **Two-tier Gemma 4 inference.** Cloud uses Vertex AI Gemma 4 26B for grading and the teacher assistant. On-device uses LiteRT-LM with Gemma 4 E2B (`react-native-litert-lm` 0.3.4, vendored Bazel rebuild from main) …