FRONTIER LLM POST-TRAINING · RL TRAINING ENVIRONMENTS
Hi, I’m Bill. I turn new ideas into AI that ships.
I’m drawn to problems nobody has cracked yet, and I build for durable, long-term impact — originating new directions, then driving them to production. Frontier-model post-training at Amazon AGI today; the AI behind autonomous MRI at Q.bio before that.
NOW · AMAZON AGI
Teaching frontier models to reason — and to act.
Post-training for frontier LLMs at Amazon AGI — where the model stops being a text predictor and starts being an agent.
RL training & verifiable rewards
Reinforcement learning for long-horizon reasoning — GRPO-style rollouts, verifier-backed rewards, and the train/eval infrastructure that turns a good idea into a repeatable model-improvement loop.
Curriculum-calibrated gyms
Synthetic environments tuned to keep a model at the edge of its ability — difficulty that scales with capability, so reward never saturates or goes all-zero, and the reasoning gains transfer beyond the training domain.
PREVIOUSLY · Q.BIO
Multimodal diffusion models for autonomous MRI.
Same instinct, earlier chapter: push frontier generative models until they actually work. I led the AI behind Q.bio’s autonomous MRI — billion-parameter multimodal volumetric foundation models and diffusion-based reconstruction — and shipped 3D models into the scanner at sub-second inference.
3000×
faster reconstruction
27×
higher image quality
9×
faster scans
/ THE ARC
From reconstructing scans to training reasoning agents.
/ PUBLICATIONS
MICCAI 2018
Efficient and Accurate MRI Super-Resolution using a Generative Adversarial Network and 3D Multi-Level Densely Connected Network
A GAN with a 3D densely-connected generator that recovers high-resolution detail from low-resolution MRI — sharper scans without more scan time.
Y Chen, F Shi, AG Christodoulou, Y Xie, Z Zhou, D Li
ISBI 2018
Brain MRI Super Resolution Using 3D Deep Densely Connected Neural Networks
3D deep densely-connected networks for brain-MRI super-resolution — the earlier study the MICCAI work built on.
Y Chen, Y Xie, Z Zhou, F Shi, AG Christodoulou, D Li
Medical Physics 2020
Fully Automated Multi-Organ Segmentation in Abdominal Magnetic Resonance Imaging with Deep Neural Networks
Deep networks that automatically delineate multiple organs across abdominal MRI — replacing slow manual contouring.
Y Chen, D Ruan, J Xiao, L Wang, B Sun, R Saouaf, W Yang, D Li, Z Fan
WACV 2020
Enhanced Generative Adversarial Network for 3D Brain MRI Super-Resolution
An enhanced adversarial network for 3D brain-MRI super-resolution (co-first author).
J Wang, Y Chen, Y Wu, J Shi, J Gee
/ OFF THE CLOCK
When I’m not training models, I’m training for marathons.
When I'm not training neural networks, I'm training for marathons. Running helps me think through complex problems and stay grounded.
3:25
MARATHON PB
5+
FULL MARATHONS
Always Building
Off the clock I tinker — a self-hosted journal, a little home lab humming with side projects, small tools I build for myself. Keeps the craft hands-on and the curiosity fed.