OntarioEdAI – The Cognitive Symbiote Platform

Version: 0.4 (Full Curriculum Database + RAG Foundation + SQLite Integration) Target: Windows tablet / OEM White-Label "Symbiote" Hardware (Offline-First) Goal: To establish a decentralized, privacy-first, on-device cognitive symbiote that adapts to the individual sentience of the student. From dynamic workbooks for elementary learners to university-level academic preparation, this is a sovereign educational ecosystem.

---

The Master Plan: R&D Roadmap & Vision

We are architecting a fundamentally new paradigm for education: a decentralized, privacy-first, on-device cognitive symbiote that adapts to the individual student. This platform completely bypasses traditional "Big Tech" cloud dependencies.

1. The Data Substrate & Knowledge Base

The AI needs verifiable grounding. We have compiled the full Ontario Grades 9-12 curriculum into a structured SQLite database and a ChromaDB vector index.

• Curriculum Database: All 30+ Ontario courses (Math, Science, English, Arts, Business, Tech, FSL, Native Languages, etc.) are deeply mapped with specific expectations.
• Custom Modules: We have expanded the baseline curriculum to emphasize Ethics, Moral Foundations, and the Evolution of Thought (tracing human spirituality from cave paintings to modern philosophy).
• Local RAG Index: Utilizes local vector storage (ChromaDB) to constrain Small Language Models (SLMs) to specific pedagogical guardrails.

2. The Dual-Engine Logic System

• Deterministic Engine: Python/SymPy and rule-based parsers handle raw computation and absolute ground truth.
• Generative Engine (Tutor): Heavily quantized SLMs (e.g., Phi-3-Mini, Llama-3 8B) translate deterministic truth into highly personalized, Flesch-Kincaid adjusted output.

3. The Symbiotic Workspace & Elementary Adaptation

• The Infinite Canvas: An interactive scratchpad (Excalidraw/Fabric.js) where students show their work using stylus input.
• Dynamic Workbooks: For younger minds (K-8), the NPU-driven SLM generates tactile, scaffolded HTML5 canvas elements (e.g., interactive fractions) instead of text-heavy queries.

4. Self-Sovereign Authentication & The Node Ledger

• Zero-Trust Security: Decentralized Identifiers (DIDs) and TEE-based AES-256 encryption ensure student data is impenetrable without local biometric/PIN access.
• Verifiable Credentials (VCs): Course mastery is stored as cryptographic badges locally, allowing seamless peer-to-peer (P2P) transfers if a student changes schools.
• Offline Mesh Network: Classrooms utilize Wi-Fi Aware (NAN) and Bluetooth Mesh to form localized peer-to-peer swarms, allowing collaboration (WebRTC) and content syncing without internet.

---

Hardware Specifications: The "Symbiote" OEM Tier

To run this platform effectively without relying on the cloud, we target a premium White-Label BOM (Bill of Materials) budget of $350–$450.

Component	Specification	Capability
Processor (SoC)	MediaTek Dimensity 8300/9300	30-40+ TOPS NPU. Runs the "Tutor" language model directly on the NPU, leaving the CPU free for rendering and OS tasks.
Memory	16GB - 24GB LPDDR5X RAM	Massive bandwidth required for fast text generation and holding large contexts in memory.
Display	12.7-inch "Paper-Matte" IPS (120Hz)	Nano-etched glass reduces eye fatigue and simulates paper friction.
Input	USI 2.0 Active Digitizer	Zero latency, tilt recognition, flawless palm rejection for the Infinite Canvas.

The True Parallel Asymmetric Multi-Agent System:

• The Watcher (0.5B Vision Model): Continuously parses the canvas for handwritten input in real-time.
• The Orchestrator: Instantly verifies mathematical/logical truth.
• The Tutor (3B-7B SLM): Resides in NPU memory, chiming in instantly with zero loading delays.

---

Honest Status Report (Current Build)

Completed & Production-Ready

• Flutter single-codebase (Windows native targeted, testing via FFI).
• Complete SQLite relational database containing 30+ Ontario courses + custom Ethics modules.
• Python rag_ingestion.py script successfully embedding curriculum into local ChromaDB for RAG.
• Riverpod state management architecture with optimized O(1) in-memory DB queries.
• Dynamic Course Overviews and detail screens reading from SQLite.

Scaffolding / Placeholders (Needs Work)

• ONNX/TFLite model bindings in Flutter (Phi-3-mini and handwriting scorer) – placeholder interfaces exist.
• The actual Excalidraw/Canvas frontend component is scaffolded but not fully integrated with a "Watcher" model.
• P2P Mesh Network protocols (Wi-Fi Aware/WebRTC) are mapped in architecture but not coded.

Supplemental Documentation

• ARCHITECTURE.md - Finer technical details on the AI logic and hardware.
• CURRICULUM_SOURCES.md - Details on the data sources, Ministry PDFs, and custom augmentations (Ethics, Grokipedia integrations, Open Educational Resources).

How to Run the App (Dev Mode)

1. Ensure Flutter is installed.
2. Run flutter pub get.
3. If running on Windows/Linux, the app automatically initializes sqflite_common_ffi to read the local assets/curriculum/ontario_curriculum.sqlite.
4. Run flutter run -d windows.

Data Pipeline (Python)

To rebuild the curriculum databases:

1. python3 parse_markdown.py (Appends new courses to the master JSON).
2. python3 migrate_to_sqlite.py (Builds the relational SQL DB for the Flutter app).
3. python3 rag_ingestion.py (Builds the ChromaDB vector index for the AI Tutor).

---

Security & Privacy

OntarioEdAI is built on a privacy-first, zero-trust architecture. Student data is encrypted at rest (AES-256) and never leaves the device without explicit user authorization.

• Vulnerability Reporting: Refer to our Security Policy for reporting guidelines.
• Dependency Management: Dependabot is enabled to ensure all Flutter and Python dependencies are up-to-date.
• CI/CD: Automated security scans (Secret scanning, Trivy, CodeQL) are performed on every push to the main branch.