Skip to content

blockchain.md

Latest commit

 

History

History
117 lines (85 loc) · 5.17 KB

blockchain.md

File metadata and controls

117 lines (85 loc) · 5.17 KB
Raw

🛡️ FlavorSnap Blockchain Architecture

FlavorSnap integrates the Stellar network and Soroban smart contracts to decentralize AI food classification governance and incentive systems. This document provides a deep dive into the blockchain layer, its components, and its role within the larger application ecosystem.

📋 Table of Contents

🚀 Why Blockchain?

Conventional AI classification services suffer from three primary issues:

  1. Centralization of Accuracy: The service provider alone determines if a classification is correct.
  2. Incentive Misalignment: Users provide data for training but don't share in the value created.
  3. Lack of Transparency: Model updates happen "behind closed doors," with no record of validation.

FlavorSnap solves these by decentralizing the verification and governance layers on the Stellar network.

🏗️ Technical Stack

  • Network: Stellar (Low-cost, high-speed, environmentally friendly).
  • Smart Contract Platform: Soroban (Rust-based, WASM-ready, efficient).
  • SDKs: Soroban Rust SDK for contracts, Stellar JS SDK for the frontend.
  • Storage: On-chain persistent storage for metadata, balances, and governance state.

⛓️ Smart Contract Ecosystem

FlavorSnap Food Registry (flavorsnap-food-registry/)

A decentralized, verifiable database of food categories and their associated metadata. This acts as the "Source of Truth" for the classification labels used by the ML model.

Model Governance (contracts/model-governance/)

Ensures that every update to the ResNet18 model used in production is transparently proposed and community-voted.

  • Proposal Lifecycle: Submission → Voting Period → Quorum Check → Evaluation → (Optional) Execution.
  • Weighting: Voting power is derived from the user's token balance (reputation-weighted).

Tokenized Incentive (contracts/tokenized-incentive/)

Provides a programmable reward layer for users who contribute high-quality data or validate classifications.

  • Vested Rewards: Long-term contributors receive tokens through vesting schedules.
  • Multi-sig Security: Minting and burning tokens require multiple admin signatures.

Sensory Evaluation (contracts/sensory-evaluation/)

The "decentralized oracle" of FlavorSnap. Users stake tokens to vouch for classification accuracy (e.g., "Is this really a Pizza?").

  • Incentive Loop: Correct evaluators earn tokens; incorrect ones lose their stake.
  • Feedback: Results are fed back into the training dataset for future RLHF (Reinforcement Learning from Human Feedback) cycles.

🔗 System Integration

graph TD
    subgraph "Frontend (Next.js)"
        UI[User UI]
        Wallet[Stellar/Freighter Wallet]
    end

    subgraph "ML API (Flask)"
        Inf[Inference Engine]
        XAI[Explainable AI]
    end

    subgraph "Blockchain (Soroban)"
        Gov[Model Governance]
        Inc[Tokenized Incentive]
        Reg[Food Registry]
        Sen[Sensory Evaluation]
    end

    UI -- "Upload Image" --> Inf
    Inf -- "Prediction Result" --> UI
    UI -- "Sign Verification" --> Wallet
    Wallet -- "Submit Evaluation" --> Sen
    Sen -- "Reward/Stake" --> Inc
    Gov -- "Approve Model Hash" --> Inf
    Reg -- "Labels" --> UI
Loading

🧩 Governance Model

FlavorSnap implements a Liquid Democracy approach (pre-alpha) where:

  1. Quorum: Proposals require a minimum percentage of total token supply to participate (e.g., 50%).
  2. Proposer Stake: A minimum amount of tokens must be staked to submit a new model proposal, preventing spam.
  3. Active Monitoring: Admins can cancel malicious proposals before they are executed.

💰 Tokenomics & Incentivization

Tokens are the fuel for the FlavorSnap ecosystem:

Action Reward (%) Unlock Schedule
Dataset Contribution 40% Immediately
Verification Staking 30% Post-Verification (7 days)
Model Improvement 20% 6-month Vesting
Governance Participation 10% Immediately

🔒 Security & Trust Model

  1. Admin Multisig: No single admin can mint tokens unilaterally. All Mint and Burn actions in TokenizedIncentive are stored in an AdminApprovals state until the threshold is met.
  2. Identity: Users are identified by their Stellar account addresses.
  3. Transparency: Every state change (token transfer, vote cast, model update) is recorded on the Stellar ledger, auditable via any block explorer.

Last updated: March 2026 For setup instructions, see docs/installation.md.