Experience

Project Manager · Polimi Data Scientists (PMDS)

Feb 2026 – Present · Milan, Italy

Lead team coordination, project execution and model development for the Maxillary Canine Impaction project.
Design an anatomy-prompted ROI attention network to predict canine impaction from panoramic radiographs.
Build a multi-task deep learning pipeline combining localization, classification and clinical risk scoring.
Apply landmark-guided feature fusion to improve interpretability and reduce ambiguity in 2D dental imaging.

Contribute as a team member to AI-driven sleep monitoring on wearable hardware.
Develop a sleep phase detection model for local inference directly on the device.
Design for 256 KB RAM and a target model size of at most 50 KB.
Optimize memory use and inference efficiency for real-time on-device sleep phase detection.

Oct 2025 – Feb 2026 · Milan, Italy

Prototyped and developed generative AI solutions for text, images, video and 3D across application domains.
Prototyped low-code MVPs for internal products.
Developed APIs and microservices to support scalable digital platforms.
Implemented CI/CD pipelines, containerization and cloud deployment workflows.
Collaborated with creative teams to integrate AI and immersive technologies into art and storytelling.
Contributed to research and innovation projects with universities and technology partners.

Jan 2022 – Aug 2024 · Kerala, India

Led a five-person AI research team, contributing to more than 10 journal and conference publications in machine learning, deep learning, deep reinforcement learning and computer vision.
Led a three-person team and an international collaboration with Dr. Dugki Min at Konkuk University on DRL algorithms for image categorization. Achieved 97.57% accuracy on MNIST, outperforming ML/DL baselines across benchmarks.
Implemented a hybrid VGG16–XGBoost gaze estimation model, achieving 98.48% accuracy across datasets with varied environmental conditions.
Guided a three-person team developing a multi-agent DRL ensemble for autonomous galaxy classification, achieving 98.55% accuracy on Galaxy Zoo.
Developed a multimodal yoga pose recognition framework combining MoveNet, ResNet50 and attention fusion. Introduced the YFA-30 dataset and achieved 96% accuracy.