Sina Shahandeh describes a research workflow in which agents receive organized information about the problem, data, models, losses, metrics and code. Hierarchical documents and retained lessons help prevent repeated mistakes and preserve context across experiments.
Sina Shahandeh uses CT-to-PET image synthesis and registration problems to show why better hypotheses matter when implementation work reaches a plateau. Agents can propose experiments and critique alternatives, but these examples do not establish clinical diagnostic validity or general scientific competence.
Sina Shahandeh combines quantitative metrics with qualitative image review and collaborative or adversarial agent critique. Misalignment caused by posture and respiration can confound results, so experimental interpretation still requires careful evaluation and human judgment.
Watch on YouTube
