San Francisco, California, United States
My Writing on Research Modelling Learnings & Best Practices: https://aurielws.github.io/writing.html 🎓 Harvard Comp Sci ’20 | YC W20 alum | Angel Investor (deep tech) | 🧑‍💻 Product Lead @ Google DeepMind Gemini (Self Evolving Agents and Systems for Gemini) Current: Product Lead @ Google DeepMind Gemini on Self Improving Agents and Modelling Spent most of my career in Modelling / AI, shipping Diffusion and Autoregressive models into products at global scale, from Gemini and YouTube to Cloud Vertex AI and Pixel. My human detection work on Pixel Real Tone earned the Grand Prix at Cannes Film Festival. Also published in Published in NeurIPS + FACCT. Before DeepMind, I co-founded a YC W20-backed algo trading startup as technical founder and studied Computer Science at Harvard. Alongside operating, I angel invest across deep tech—supporting founders building at the edge of what’s possible.
Currently: Teaching Gemini some agentic coding skills (experimental SWE Agents, harness/ scaffolding system optimization, Pretaining and RLHF focused approaches, etc) 👩🏾‍💻 Check out Gemini 3.0 on AI Studio, Gemini CLI, Jules 🦑 , and Antigravity H2 2024: Making Gemini work well for real world use cases :) + Alignment and High quality visual outputs across Deepmind models (NanoBanana, Veo, etc). H1 2024: Responsible Generative AI Visual models from pre-training to deployment. Lots of evals... Checkout Imagen 3 in Bard and the imagen 3 tech report here: https://storage.googleapis.com/deepmind-media/imagen/imagen_3_report.pdf Check out Gemini native image generation on ai studio now :)
Core work was mainly co founding internal Gen AI Signals product. Models are used across Google LLM products, including Bard, Cloud, MakerSuite, and Google Slides. Also still did some human detection / Perception stuff when I had time
Core work -- Human Detection / Computer Vision Efforts. Emphasis on making sure the models x-google work for various human attributes (skin tones, ages, genders depictions, etc). Efforts improved human detection for products like Google’s Image Search, Pixel Camera auto white balance ML (Pixel Real Tone), and other google computer vision based products.
Not taking new clients until further notice.
Co-founder and CEO of college startup StrattyX. StrattyX did YC, Harvard ilab, Tech Crunch Disrupt, Grace Hopper, raised a bunch of money and some other cool stuff but I mainly learned how to build products from 0 to 100 under time and under budget and how to grokk various technical stacks and problems with ease.