Maximiliano Levi optimizes inference for audio models at ElevenLabs. His projects include a CPU text-to-speech engine built with C++ and ggml, a ray tracer with CUDA support, and the open-world game Project Hedra.

Unclaimed profile

Maximiliano Levi optimizes inference for audio models at ElevenLabs. His projects include a CPU text-to-speech engine built with C++ and ggml, a ray tracer with CUDA support, and the open-world game Project Hedra.

Career

Inference Lead / Research Engineer at ElevenLabs until PresentWorks on inference and optimization for ElevenLabs' AI systems. Publicly credited for inference and optimization work on ElevenLabs Scribe. Based in the ElevenLabs engineering and research organization.
maxilevi.com

Writing a CPU only text to speech engine

Maximiliano explains his C++ and ggml port of VITS, including profiling, convolution reformulation and numerical tradeoffs for running speech synthesis on CPUs.

Open ↗
maxilevi.com

Generating procedural worlds in Project Hedra

Maximiliano walks through the storage, terrain generation and rendering techniques behind his open-world game, including compact voxel data and mesh optimization.

Open ↗

Maximiliano's Collections

Maximiliano's Lists

In these lists

Cosigns Given