James Ray

I help AI research teams hill-climb and ship frontier models from 0-1. And I prototype harnesses and control layers to turn emerging GenAI capabilities into relatable product experiences. Currently the Lead PM for Multimodal Research on Mustafa Suleyman's Superintelligence Team at Microsoft AI. Responsible for 0-1 foundation model research in image gen, video gen, code gen, world models, and speech understanding/synthesis. Expanded scope to own all Media Gen products on Copilot, including OpenAI model deployment + scaffolding. As a hands-on Research PM, I contribute to technical decisions and ablations across the research lifecycle, e.g.: - annotation experiments and evals (e.g., VLM captioner alignment) - quality evaluation in nondeterministic systems (both human and machine evals) - post-training and alignment (SFT and RL), model behaviors - system prompting and agentic harnesses - safety guardrails operating at global scale Through this process, I've helped research teams hill-climb 7+ models to the the top-5 of LMArena and the top-1 of key benchmarks. I shipped MAI's first in-house models: Voice-1, Transcribe-1, Image-1, Image-2, Image-2e (efficient) I've also led significant turnarounds in response quality and model behavior in key domains for our global conversational AI product. Part of my job is to predict where the technology is going next. To do that, I partner with research leadership to define research roadmaps, and I prototype novel problem formulations and harnesses to deliver the next set of breakthroughs. --- At Netflix I built Studio GenAI from 0-1 globally, working with filmmakers, game devs, and AI researchers to train and fine-tune text, image, and video generation models for film and games. I partnered closely with creators - including Academy Award and Emmy winners/nominees - to pioneer artist-driven methods for GenAI alignment, contro