Shard

The inspiration for Shard came from observing two key problems in the modern computing landscape: Wasted Computing Power: Millions of high-performance Macs sit idle for hours each day, their powerful Apple Silicon chips underutilized while consuming electricity. Wasted Computing Power: Millions of high-performance Macs sit idle for hours each day, their powerful Apple Silicon chips underutilized while consuming electricity. Centralized AI Infrastructure: Running AI inference and compute-intensive tasks typically requires expensive cloud infrastructure or centralized services, creating bottlenecks and single points of failure. Centralized AI Infrastructure: Running AI inference and compute-intensive tasks typically requires expensive cloud infrastructure or centralized services, creating bottlenecks and single points of failure. I asked myself: What if I could create a decentralized marketplace where anyone could contribute their idle computing power and earn cryptocurrency, while others could access distributed compute resources at a fraction of traditional cloud costs? The name "Shard" reflects the core innovation—breaking large computational jobs into tiny micro-task fragments that can be processed independently across a distributed network, similar to how blockchain shards distribute data and computation. Shard is a decentralized compute network built on the Sui blockchain that transforms idle Mac computers into a distributed supercomputer. Here's how it works: For Job Submitters (via Web Interface) Submit Computational Tasks: Upload a CSV file with 100 sentences for AI content moderation or provide parameters for Monte Carlo simulations Automatic Fragmentation: The system breaks your job into tiny independent fragments (1 sentence per fragment) Privacy-First Encryption: Each fragment is encrypted using Sui Seal threshold encryption Decentralized Storage: Encrypted data is stored on Sui Walrus (decentralized blob storage) On-Chain Job Posting: The job and all fragments are published to the Sui blockchain with USDC bounties (0.01 USDC per fragment) Real-Time Progress: Watch as workers around the world claim and process your fragments Automatic Result Aggregation: Get your completed results delivered in the original order For Computing Providers (via macOS App) Beautiful System Monitor: Track CPU, GPU, and memory usage in a Robinhood-like interface One-Click Worker: Click "Start Worker" to begin earning USDC AI-Powered Chat: Built-in Gemma 3-4B chatbot for local AI assistance Auto-Discovery: The app polls the blockchain for available work Secure Processing: Downloads encrypted fragments, decrypts locally using Seal Local AI Inference: Runs Gemma AI model to classify content (safe/unsafe) Upload Results: Encrypts and uploads results to Walrus, submits completion on-chain Instant Payment: Receives 0.01 USDC per completed fragment (sponsored transaction—zero gas fees) Multi-Worker Support: Spawn multiple worker windows to process tasks in parallel Key Features Zero Gas Fees for Workers: All transactions are sponsored by job creators Privacy-Preserving: Workers never see raw sensitive data—only encrypted fragments Fault-Tolerant: Uncompleted fragments are automatically re-queued Native Performance: Leverages Apple Silicon's Metal acceleration for AI inference Real-Time Transparency: Both submitters and workers see live progress updates I built Shard solo over 15 hours during the hackathon, integrating cutting-edge blockchain technologies with native macOS development. Tech Stack Frontend (Next.js 14 + React + TypeScript) UI Framework: shadcn/ui components with Tailwind CSS for a minimalist, dark-mode-first design Animations: Framer Motion for smooth transitions and progress indicators Real-Time Updates: Polling-based live fragment status tracking Features: Job submission portal, AI inference configuration, Monte Carlo simulation setup Backend (Node.js + Express + TypeScript) API Server: RESTful endpoints for job management, fragment lifecycle, and wallet queries Sui Integration: @mysten/sui SDK for blockchain interactions and USDC transfers Walrus Client: Custom implementation for uploading/downloading encrypted blobs Seal Encryption: @mysten/seal SDK for threshold encryption with session keys Job Orchestration: Automatic fragmentation, encryption, and on-chain publishing macOS App (SwiftUI + Combine) System Monitoring: Real-time CPU/GPU/Memory tracking using IOKit and Mach APIs AI Integration: Embedded llama.cpp server running Gemma 3-4B-it-q4_0.gguf model Metal Acceleration: Native Apple Silicon GPU support for AI inference Worker Engine: Multi-threaded fragment discovery, claiming, processing, and completion Multi-Window Architecture: Spawn independent worker windows for parallel processing Live Dashboard: Beautiful glassmorphic UI with animated charts (inspired by Apple's design language) Blockchain (Sui + Move) Sui Network: Testnet deployment for job and fragment state management Smart Contracts: Move modules for access control policies (Seal) Walrus Storage: Decentralized blob storage with erasure coding Seal Encryption: Threshold encryption with programmable access control Sponsored Transactions: Job creators pay all gas fees for workers Development Highlights 1. Micro-Task Fragment Protocol I designed a novel fragmentation system where each sentence becomes an independent TaskFragment with: Unique fragment ID and job ID Encrypted data blob stored on Walrus USDC bounty (0.01 per fragment) Status lifecycle: pending → claimed → completed 2. Sui Seal Integration I implemented end-to-end encryption using Seal's threshold encryption: Generated session keys with TTL for time-limited access Created Move access policy contracts (seal_approve) Encrypted fragments using AES256GCM Workers decrypt using session keys without exposing raw data 3. llama.cpp Embedding I bundled the Gemma AI model and llama-server binary directly into the macOS app: Built static llama-server to avoid dynamic library issues Spawned server as subprocess with Metal GPU acceleration Implemented streaming HTTP client for real-time inference Created beautiful chat UI with gradient avatars and suggestion chips 4. Multi-Worker Architecture I built a sophisticated worker management system: @Environment(\.openWindow) for spawning new worker windows Shared WorkerService instances with independent worker IDs Competitive claiming (first worker to claim gets the fragment) Real-time USDC balance tracking across all workers 5. Production Deployment Frontend: Deployed to Vercel (https://shard-alpha.vercel.app) Backend: Deployed to Vercel with Cloudflare Tunnel for local testing CORS: Dynamic subdomain support for *.trycloudflare.com Config Management: Centralized API URLs for easy environment switching Blockchain & Cryptography Sui Blockchain: I dove deep into the Sui ecosystem, learning about Move smart contracts, sponsored transactions, and how to structure on-chain state for a compute marketplace. Sui Walrus: I implemented decentralized storage for both job inputs and results, learning how erasure coding provides redundancy without centralized servers. Sui Seal: I mastered threshold encryption and programmable access control, enabling workers to decrypt job data without ever exposing it to centralized parties. BCS Encoding: I learned about Sui's Binary Canonical Serialization for efficient data encoding. Native macOS Development I built a production-grade SwiftUI application with real-time system monitoring (CPU, GPU, Memory). I integrated llama.cpp and Gemma 3-4B for local AI inference using Apple Silicon's Metal acceleration. I learned advanced Swift concepts: Combine framework, @Published properties, URLSession streaming, and multi-window architecture. I mastered IOKit for low-level hardware metrics and Mach APIs for system-level monitoring. Distributed Systems Design I implemented a micro-task fragment protocol where jobs are broken into independent units that can be processed in parallel. I designed fault-tolerant execution with fragment re-queuing if workers go offline. I built a polling-based worker discovery system that scales to multiple simultaneous workers. I learned about eventual consistency and how to handle race conditions in distributed claiming. Full-Stack Integration I connected React/Next.js frontend → Express.js backend → Sui blockchain → macOS worker app in a seamless end-to-end flow. I implemented real-time progress tracking with polling-based live fragment status updates. I managed CORS, API design, environment configuration, and production deployment. I debugged complex multi-component issues (e.g., USDC transfers, fragment completion, UI state synchronization). Performance Optimization I optimized SwiftUI rendering by removing continuous animations and using .drawingGroup() for Metal acceleration. I reduced CPU usage from 50% to <5% by simplifying chart rendering and using efficient update intervals. I implemented smart batching for blockchain queries to minimize network overhead. Short-Term (Next 3 Months) Mainnet Launch: Deploy to Sui mainnet with real USDC incentives More Task Types: Support image processing, video transcoding, and scientific simulations Advanced Reputation System: Track worker reliability and completion rates with on-chain NFTs Mobile Support: Build iOS/Android worker apps to expand the provider network Job Verification: Implement cryptographic proofs of correct computation Long-Term Vision Cross-Chain Support: Integrate with Ethereum, Solana, and other L1s for broader adoption Specialized Hardware: Support GPU-intensive tasks (3D rendering, AI training) with CUDA/ROCm Enterprise Partnerships: Offer B2B solutions for companies needing distributed compute DAO Governance: Community-driven protocol upgrades and fee structures Carbon Credits: Reward providers for using renewable energy sources Technical Improvements WebSocket Real-Time Updates: Replace polling with push notifications for instant updates Smart Contract Upgrades: On-chain fragment verification and slashing for malicious workers Result Validation: Implement consensus mechanisms (multiple workers verify the same fragment) Benchmarking System: Automatically test worker capabilities and assign appropriate tasks Dynamic Pricing: Market-driven bounty pricing based on task complexity and demand Shard demonstrates that it's possible to build a fully functional decentralized compute network in a single weekend. By leveraging Sui's innovative blockchain architecture, Walrus's decentralized storage, and Seal's privacy-preserving encryption, I created a platform that turns idle Macs into a global supercomputer—while ensuring privacy, fault tolerance, and instant micropayments. This is just the beginning. The future of computing is decentralized, and Shard is proof that we can build it today.