Add Vision-O3: Edge-AI Model Registry & Verifiable CV Datasets

[Keerthivasan-Venkitajalam]

Project Overview

Vision-O3 is a decentralized edge-AI model registry and verifiable computer vision dataset management system built on Akave O3 storage.

Problem Statement

Edge devices (Raspberry Pi, IoT) currently pull model weights from centralized hubs (HuggingFace, AWS S3), creating:

• Single points of failure and vendor lock-in
• Difficulty versioning and verifying large training datasets
• Lack of cryptographic verification for model integrity
• Infrastructure overhead for edge deployments

Solution

Vision-O3 provides:

• Decentralized Storage: Model artifacts (.pt, .onnx, .tflite, .h5) and datasets stored immutably on Akave O3
• CID Verification: Content-addressed storage ensures model integrity at the edge
• FastAPI Gateway: Lightweight REST API with JWT authentication and role-based access control
• Python SDK: Drop-in client library for edge devices with automatic verification and caching
• Version Management: Full version history and metadata tracking for reproducibility

Deliverables

• FastAPI Gateway service with complete REST API
• Python Client SDK for edge devices
• Akave O3 integration layer with retry logic
• Metadata store (PostgreSQL/SQLite) for fast discovery
• Docker containerization for easy deployment
• Comprehensive documentation and examples

Validation Goals

• Prove Akave O3 is suitable for edge-AI model distribution
• Demonstrate decentralized storage for reproducible ML workflows
• Showcase content-addressed verification for model integrity
• Validate Akave O3 for high-bandwidth binary artifacts

Alignment with Akave ICP

This project validates Akave O3 as a backend for:

• High-bandwidth binary artifacts (model weights, datasets)
• "Warm" data use cases for AI training and edge deployment
• Verifiable AI systems requiring cryptographic integrity
• Cost-efficient alternative to centralized cloud storage

Files Included

• vision-o3/PROPOSAL.md - Detailed project proposal
• vision-o3/PLAN.md - Complete implementation plan with milestones

Contributor

Name: Keerthivasan S V
Background: Computer Vision (YOLOv11), Edge AI (Raspberry Pi), FastAPI, Docker
Previous Work: IoT-based pipe inventory management, CNN-based medical imaging

---

Looking forward to feedback and collaboration opportunities!