Chris
Miles

I'm an applied mathematician working in numerics and AI compute, with roots in dynamical systems.

My academic training was in nonlinear dynamics — PhD in physics and MS in applied math at the University of Michigan. My thesis applied PDE-constrained optimal control to fluid mixing. In addition, I worked on other fluid dynamics problems in active matter and acoustic droplet vaporization.

Out of graduate school I spent two years as a data scientist and machine learning engineer at a fintech startup, building a suite of time-series forecasting models — ARIMA-like models, Markov-chain models, and time-delay embedding methods from nonlinear dynamics.

Since 2021 I've been at Quadric, a custom AI inference processor startup, where I lead numerics and quantization. I take neural networks (CNNs and LLMs) down to int4/int8 on the chip's arithmetic with minimal accuracy loss, and build the kernels, error analysis, and validation tooling that make that possible. That analysis sometimes lands in the hardware itself: I quantified the error cost of different design choices in the chip's fp16 multiply-add, and wrote a custom fx32→fp16 conversion instruction in Verilog.