Aakash Varma Nadimpalli

Led the development of deep learning and AI-driven computer vision models, enhancing image segmentation and classification for real-world applications. Designed and optimized semi-supervised learning (SSL) algorithms that improved automatic image annotation accuracy, enabling enhanced decision-making with limited labeled data. Developed and fine-tuned predictive AI models, leveraging novel loss functions to enhance model generalization and performance on large-scale datasets. Implemented MLOps practices and CI/CD tools (GitHub Actions) for automating model training, testing, and deployment pipelines, ensuring reproducibility and consistent quality. Built and deployed AI systems for deepfake detection on cloud platforms (GCP, AWS) to enable scalable inference and model serving across diverse environments. Applied deep understanding of statistical learning methods to improve model performance and interpret complex patterns in visual data. Spearheaded the integration of Generative AI (GenAI) methodologies, incorporating diffusion models and adversarial training techniques to improve data efficiency and model robustness. Queried and mined large-scale datasets using Snowflake, integrating Python and SQL-based pipelines to support cloud-scale analytics workflows. Designed and implemented scenario modeling, clustering, risk scoring, and anomaly detection frameworks to inform strategic tradeoff decisions. Collaborated with cross-functional stakeholders to gather business requirements, promote insight adoption, and translate analytics into strategic recommendations. Practiced Agile methodologies in collaborative sprints, accelerating model iteration, deployment, and feedback loops. Conducted rigorous benchmarking and performance evaluation of AI/ML models, refining architectures to optimize accuracy and reduce false positives. Established streamlined workflows for autonomous training, validation, and deployment of models, ensuring dynamic learning and scalability. Stayed at the forefront of AI advancements, experimenting with reinforcement learning, prescriptive analytics, and GenAI techniques to drive continuous model improvement.

Conducted research to develop scalable and transferable deep learning models for domain adaptation and detection of facial manipulations, leveraging advancements in generative AI. Worked on harnessing unlabeled data for improvement and generalization of deep learning models, ensuring robustness across diverse datasets. Developed generalizable deep learning models for preventing deepfakes across social media platforms, enhancing AI-driven security measures. Published and presented research at leading AI conferences, contributing to advancements in deepfake detection, AI security, and explainable machine learning. For more information, refer the publications.

Worked as graduate teaching assistant for machine learning, deep learning, advanced artificial intelligence, data science, NLP, digital image processing and image analysis courses. My duties included delivering lectures, facilitating discussions, and supporting students in their understanding of the fundamental concepts and applications within the field of machine learning.

Defined and planned computer vision data collection efforts to further biometric authentication and spoof detection research. Executed data collection efforts using custom-built data collection apps and methods to ensure high-quality and diverse datasets. Classified attributes for collected data to optimize training algorithms and improve model performance in spoof detection. Developed spoof attack methods and generated synthetic spoof data to enhance adversarial robustness and improve detection accuracy.

Doctor of Philosophy (Ph.D.) - Computer Science

Master of Science - (Electrical Engineering)