I'm co-founder and CEO of Avartha. Today, real-time AI such as voice runs on small models because frontier models are too slow. We're rebuilding the inference stack so that real-time applications can use the most capable models.

Avartha spun out of the Systems for AI Lab at Georgia Tech, where I completed my CS PhD with Prof. Alexey Tumanov. Our scheduling and parallelism work, including chunked prefills from Sarathi-Serve, runs in vLLM, SGLang, and TensorRT-LLM. Before that, I built training infrastructure at Microsoft Research and Qubole.