Tensor Royale

Screenshots

Problem Statement

Tensor Royale is an incentivised AI collaboration platform. It allows for a permissionless upload of training data and computational models. It incentivises creation and sharing of open source models and data, by providing user-submitted challenges & incentives. It allows for independent model evaluation while taking care of data consistency. We also provide the ability to deploy these models on-chain using STARKWARE to prove the validity of execution.

Solution

Our technology stack includes partner technologies from Filecoin, Avail, Nethermind/Starkware and Dynamic. Frontend is build with React, using ethers and Dynamic wallet for a hybrid web2/web3 login, which fits our target group (AI researchers) who might not be entirely familiar with web3. All file hosting on frontend is provided by Filecoin, which provides efficient, long-term and infinitely scalable storage. Data storage is paid by the users, based on per-use basis. On the backend we leverage SHARP (Starkware's prover) to efficiently execute and batch a large number of transactions, ensuring sufficient scalability for unique use-case while ensuring the integrity of execution and user self custody of data and funds. Paired with Avail, a high performance data availability solution, this allows us to offer the best user experience.Video uploaded here (last minute export issues😅): https://k00.fr/8lh1f33r | password: 901376