Summary
Sasha Goldshtein is a Principal Software Engineer with 13+ years of experience, currently leading cross-functional efforts at Google to improve factuality for Gemini across text, images, video, and long-context models. He blends deep applied ML leadership—covering pretraining data interventions, SFT, RM, RLAIF and RLHF—with hands-on measurement work in both human and automatic factuality classifiers. Previously a CTO, international speaker, trainer and author, he wrote "Pro .NET Performance" and built a global training practice that delivered courses and conferences to thousands. A prolific open-source contributor, he’s helped projects like async-profiler (adding jattach and tooling) and BPF-based tracing labs that demonstrate real-world debugging and profiling techniques. Known for turning research into production features (e.g., Google Duplex and enhanced Search/Maps data quality), he combines tooling depth in systems and profiling with cutting-edge multimodal LLM research. Based in Kfar Sava, Israel, he brings a rare mix of enterprise-scale product delivery, developer education, and low-level performance expertise to AI safety and factuality challenges.
13 years of coding experience
18 years of employment as a software developer
B.Sc. Computer Science, B.Sc. Computer Science at The Hebrew University of Jerusalem
Leyada
Hebrew, English, Russian