APEX Evening: The State of High-Performance Inference, with Omer Shlomovits (MoonMath.ai)
AI inference and serving performance depends on how well hardware, software, and model architecture work together. Omer Shlomovits, founder of MoonMath.ai, takes a low-level systems approach, optimizing across these layers to achieve state-of-the-art end-to-end performance.
The State of High-Performance Inference
This talk surveys the principles and techniques behind high-performance inference for frontier open-weight LLMs, drawn from the trenches of optimizing and serving these models in production. We'll cover the key optimizations at the model level, in serving engines, and in low-level libraries — with a particular focus on kernel optimization. We'll also look ahead: the emerging role of AI agents, and where inference hardware is headed.
Speaker Omer Shlomovits — Founder & CEO of MoonMath.ai, building optimized, multi-hardware inference software. Deep-tech entrepreneur (1x exit), Psagot, 8200, 81.
APEX hosts high-signal conversations for Israel's AI builders — no panels, no networking theater.