Generative AI Researcher, interested in Robotics & Reinforcement Learning
I recently completed an M.Sc. in Computer Science researching generative AI, and I'm now interested in exploring robotics and reinforcement learning, bringing a constraint-driven, systems-level way of thinking to planning and decision-making for embodied agents.
I'm a generative AI researcher. My M.Sc. work tackled 3D-consistent text-to-image synthesis, treating multi-view consistency as a constraint-satisfaction problem, and I'm now interested in exploring robotics and reinforcement learning, where I'd like to apply a similar constraint-driven approach to planning and decision-making in physical, embodied systems.
I've had the chance to look at visual collection, planning, and analysis from two different directions: operationally, during my time in the army, and academically, during my M.Sc. research. Before returning to research, I spent several years leading technical and operational work in the defense technology sector: building production software for intelligence systems, planning and coordinating complex multi-team operations, and operating high-performance, safety-critical systems that demanded precise, real-time decision-making under tight constraints. That background shapes how I approach research, breaking down ambitious goals into rigorous, verifiable steps.
My M.Sc. thesis work, the focus of my research to date:
Generating geometrically consistent reflections in mirrors by combining geometry estimation, projection, and dual-mask diffusion-based inpainting. M.Sc. thesis work; paper currently under review.
Unit 8200, Israel Defense Forces
Built and shipped production software for visual-intelligence systems within a national signals and technology intelligence organization; designed high-throughput, high-accuracy data pipelines.
Defense & national security technology sector
Planned and led large-scale, multi-team operations: decomposed high-level objectives into detailed execution plans, coordinated across units, and adapted plans in real time under tight accuracy and timing constraints.
Defense & national security technology sector
Trained on and operate complex, high-performance, safety-critical systems requiring precise, high-stakes decision-making in real time. Led a team specializing in visual-intelligence collection operations (2014–2020).
Research under Prof. Ohad Fried and Prof. Yacov Hel-Or: generative models for text-to-image synthesis, focused on 3D consistency within generated imagery, requiring the model to reason about scene geometry and satisfy constraints across multiple views.