DripChain

Inspiration

Traditional resale platforms are bogged down by high fees, slow banking settlements, and a lack of transparency. We wanted to create a platform that leverages the Solana blockchain to enable instant, low-fee, peer-to-peer global transactions. However, we didn't just want to build a crypto marketplace; we wanted to solve the "discovery" problem in vintage fashion. Our goal was to combine the security of decentralized finance (DeFi) with the user experience of modern e-commerce, powered by intelligent visual recommendations.

What it does

DripChain is a Web3-native web app where users can buy and sell clothing directly using Solana (SOL).

Wallet-Based Identity: No emails or passwords. Users sign in instantly by connecting their Solana wallet (Phantom, Solflare, Brave).

Seamless Marketplace: Users can list items by uploading photos and details (condition, size, price). These listings are instantly viewable by the community.

Transparent Transactions: Purchases are executed via blockchain transactions, creating an immutable ledger of sales that prevents fraud and ensures transparency. Sold items are permanently marked, preserving the history of the marketplace.

AI Visual Recommendations: When a user "saves" an item they like, DripChain analyzes the image and automatically recommends visually similar clothing from other sellers using a custom machine learning model.

How we built it

We built DripChain as a full-stack Next.js application integrated with a Python-based ML backend.

Frontend: Built with Next.js and TypeScript for a responsive, type-safe user interface.

Blockchain Integration: We used the Solana Wallet Adapter to handle connection states and transaction requests for Phantom, Brave, and Solflare wallets.

Backend & Storage: We utilize Supabase for real-time database needs and Pinata (IPFS) for decentralized image storage.

Machine Learning Engine:

Training: We trained an EfficientNet-V2 Convolutional Neural Network (CNN) on the DeepFashion dataset from Kaggle.

Inference: Built with PyTorch and served via a Flask API.

The Algorithm: The model converts listing images into feature vectors. When a user saves an item, we perform a k-nearest neighbors search using Euclidean distance to find and recommend the most visually similar items currently for sale.

Challenges we ran into

Vector Similarity Search: Implementing the Euclidean distance logic efficiently was a challenge. We had to ensure that comparing the feature vectors of saved items against the entire database didn't slow down the user experience.

Integration: Bridging the gap between our Next.js frontend and the Python/Flask ML backend to ensure recommendations are loaded seamlessly.

Accomplishments that we're proud of

Successfully integrating a custom Deep Learning model into a Web3 application—something rarely seen in dApps.

Creating a fully functional P2P payment loop on the Solana Devnet.

What's next for DripChain

Smart Contract Escrow: Implementing an escrow system where funds are held in a smart contract until the buyer confirms delivery.

Hybrid Filtering: Improving the recommender system to weigh text-based attributes (like Brand or Size) alongside the visual image features.

Mobile App: Porting the platform to React Native for a mobile-first experience.

Built With

• efficientnet-v2
• flask
• next.js
• pinata
• python
• pytorch
• solana
• supabase
• typescript