Obin Labs

Obin Labs represents my most ambitious project to date - a distributed computing platform that democratizes access to advanced AI by enabling large language models to run across networks of ordinary smartphones. This innovation addresses two critical challenges in AI deployment: the centralization of computing power and the environmental impact of large data centers.

The core technology involves a novel approach to model sharding that I developed using PyTorch. Unlike traditional sharding techniques, my method dynamically allocates computation based on device capabilities and network conditions, allowing even entry-level smartphones to contribute meaningfully to inference tasks. The system uses a custom-built peer discovery and coordination protocol written in Go that achieves remarkable efficiency even on unstable mobile networks.

One of the most significant technical achievements was developing a differential privacy mechanism that ensures user devices never expose sensitive data while still contributing to the shared computing resource. This required implementing advanced cryptographic protocols and a lightweight container solution optimized for mobile operating systems.

The AWS component provides the coordination layer and fallback computing resources when the peer network cannot meet demand thresholds. My implementation uses a serverless architecture to minimize costs during varying load conditions.