Working with prospective and existing customers on LLM inference engine optimizations, text-to-speech (TTS), image diffusion models, protein folding (+MSA), synthetic data generation, etc.

Developing ML examples for modal.com/examples, incl:

• Train an LLM from scratch using torch on Shakespeare text

• Protein Folding with ESM3 & boltz-1

• Browserman Agent

* Solving #1 complaint in hearing industry — Difficulty hearing in noisy social situations.

* Raised money from YCombinator, Xoogler, and National Institute on Aging (grant).

* Developed team of contractors across hardware, machine learning, acoustics, and hearing difficulty.

* Collaborated with Stanford, John Hopkins, & University of the Pacific on clinical benefits.

* Designed an acoustic ray-tracing engine to generate tens of gigabytes of highly accurate acoustic training data for training our deep learning models using 16 AWS GPUs simultaneously.

Developed a novel patented thesis for solving the cocktail party problem - Enhance voices nearby patients and suppress ones further away.

* Method: Develop a synthetic acoustic dataset of voices from different distances and train a deep learning model on it.

* Developed an acoustic ray tracing engine from scratch that could generate binaural room impulse responses (BRIRs) 80% as accurate as ones from the real world using 16 GPUs on AWS.

* https://www.youtube.com/playlist?list=PLTJbNpfeLPSgiFzJaD6fzFT20L6Qf_ZBc

Clinical Research into the benefits of our technology including increased interest in socializing, noise tolerance, and word recognition in noise.

* Set up audiology booth for 2-channel testing of QuickSIN and Acceptable Noise Level (ANL) with two researchers.

* Researchers found 2-3X better noise tolerance with our technology.

* Applied & Awarded A2 Pilot Grant from the NIA & John Hopkins University for our research.

* Collaboration with Dr. Jiong Hu at University of the Pacific & Professor Fitzgerald at Stanford