The widespread adoption of machine learning models across various application domains has been driven by easier access to data. To address concerns related to data privacy, security, and integrity, collaborative learning paradigms such as Federated Learning and Split Learning have emerged. The transition to these techniques has been further accelerated by the growth of data marketplaces and collaborative learning platforms. A key challenge, however, remains in providing an infrastructure that enables secure data sharing. A decentralized, transparent, and privacy-preserving storage layer can not only address these issues but also enhance the efficiency of such platforms. In this work, we propose Blocklet, a purpose-built storage system compatible with Machine Learning Operations (MLOps) guidelines. Blocklet leverages Web3 technologies: it uses IPFS for persistent storage, the Ethereum blockchain to maintain tamper-proof records, and PGP encryption to ensure data security. The system architecture consists of three layers: the Blockchain layer, the Storage layer, and the Frontend layer. The Storage layer, which includes data nodes and admin nodes, forms the core of the design and, together with smart contracts, provides a complete audit trail of data updates, enhancing the platform’s resilience. We conducted experiments to measure the latency and throughput of read and write operations in Blocklet, and the results demonstrate its efficiency as a storage platform.