Autimate AI & AR Ecosystem — Feature Modules
Five AI Modules, One Ecosystem
Each module is designed to work independently or together — creating a comprehensive therapeutic environment grounded in science.
AR & Real-World Learning
Point the camera at everyday objects and watch them come alive with 3D AR overlays. Powered by on-device TFLite inference and ar_flutter_plugin_2, children hear object names in Arabic or English — bridging the gap between the physical and digital world.
- Real-time object detection
- Bilingual labels (AR / EN)
- Interactive 3D GLB models
Emotion Recognition Game
Our Swin Transformer model, exported as an optimized ONNX runtime, detects 7 facial emotions in real-time: joy, sadness, anger, fear, surprise, neutral, and disgust — each with confidence scores and probability distributions.
- Swin Transformer ONNX model
- 7 emotions with confidence scoring
- Per-child emotion summaries
Play-Based Therapy
Four game types — Memory Match, Puzzle, XO (Tic-tac-toe), and Object Identification — each with a full session lifecycle tracking correct answers, wrong answers, timeouts, and attempts. Items include image choices, audio choices, and matching pairs.
- 4 game types with 6 item formats
- Session lifecycle & success tracking
- Adaptive difficulty levels
AI Chatbot Assistant
A 6-stage RAG pipeline: Query Rewriting → Dense Retrieval (E5 multilingual embeddings) → BM25 Sparse Search → Reciprocal Rank Fusion → Cross-Encoder Reranking → Contextual Compression. Powered by Gemini 2.5 Flash Lite with child-profile personalization.
- Personalized to child's autism level & goals
- Bilingual Arabic / English support
- ChromaDB + BM25 hybrid retrieval
Smart Insights & Reports
AI-generated weekly Markdown reports powered by Gemini. Each report includes game performance snapshots, emotion trend analysis, and actionable recommendations — all version-tracked per child for reproducible clinical assessments.
- Weekly AI-generated reports
- Game + emotion analytics snapshots
- Prompt/response token tracking
