Akhil Gorugantu

Graduate Research Assistant @The Research Foundation for SUNY

Buffalo, NY, US

Currently working on developing reward-based control selection methods to improve fine-grained controllability in image and video editing models.• Dyadic Interaction & Strategy Reasoning: Developed a video understanding framework to analyze behavioral patterns (strategies) rather than semantic content; leveraged LLM priors to automatically score the completeness and fidelity of complex human interactions.• FERMI: Leveraging generative priors to perform 3D-consistent part editing within 2D imagery, ensuring geometric plausibility and physical consistency for precise scene control.• Multi-modal Video Fusion & Segmentation: Engineered a dual-stream vision pipeline fusing White Light and Near-Infrared video by fine-tuning SAM3 for pixel-level alignment, achieving a 2.5x increase in Signal-to-Background Ratio (SBR).• Audio-to-Insight Summarization Pipeline: Built an end-to-end clinical trauma severity prediction system using LLM few-shot prompting for Prehoc Trauma Prediction. The system achieved 83.4% triage accuracy (40% Overtriage and <5% Undertriage), outperforming human experts and improving real-time human decision-making by 2.57x.

JNTUH College of Engineering Hyderabad

Bachelor of Technology - BTech, Computer Science

University at Buffalo

Master's degree, Computer Science