Daniel Bump is a Google research engineer working on image and video generation, computer vision, and language models. His product portfolio includes Google Photos storage management, Veo image-to-video generation for Google Ads, and 3D-aware photo recomposition.
Bump studied at the Georgia Institute of Technology from 2017 to 2021 before building his career at Google. His visual-generation interests include super-resolution, image editing, and preserving fine details. Google’s Auto frame technology, which he includes among his projects, estimates a photograph’s three-dimensional geometry and uses latent diffusion to generate content revealed by a changed camera angle.
At AI Engineer World’s Fair 2026, Bump and fellow Google / YouTube Ads speaker Preetika Bhateja examined how advertising agents can be evaluated against creative accuracy, brand safety, and other production requirements. Bump’s specific priorities include:
- Build dependable tools first. Optimize the agent’s underlying capabilities before scaling evaluations; critique agents and remediation loops can address remaining limitations.
- Start small, then formalize. Inspect early outputs directly, identify recurring weaknesses, and build curated golden datasets around representative tasks and negative cases.
- Evaluate patterns, not anecdotes. Because generative outputs vary between runs, measure repeated failure modes instead of rewriting prompts around isolated mistakes.
- Make launch readiness measurable. Improve datasets, rating guidance, agent behavior, and tools together, while distinguishing acceptable regressions from failures that should block release.
