Supervisor: Mr. Siddarth Malreddy, Tech Lead Manager & Mr. Ishan Nigam, Senior ML Engineer, Uber

• Augmented uLabel with deep learning object tracking for automated RGB/IR video annotation, reducing manual effort
• Deployed XGBoost anomaly detection in human-in-the-loop tracking validation for frame-level annotation accuracy

Tech Stack: PyTorch, Python, C++

Supervisor: Dr. Guillaume Sartoretti, Assistant Professor, MARMoT Lab, NUS

Vision-Attention-Driven Autonomous Navigation with Semantic Understanding

• Enhanced CogniPlan with cross-attention between frontier and node embeddings to enrich node representations
• Incorporated Visual Navigation Transformer (ViNT) to capture semantic context for adaptive exploration strategies

Uncertainty-Guided Path Planning via Conditional Layout Prediction

• Architected CogniPlan, integrating Wasserstein GAN-based conditional inpainting model and graph attention network for uncertainty-aware navigation
• Achieved up to 17.7% shorter exploration paths and 3.9% improved navigation efficiency over state-of-the-art baselines across 100+ unseen maps, using a lightweight model with fewer than 0.35M parameters enabling real-time CPU inference

Tech Stack: ROS Noetic (with Gazebo, Rviz), PyTorch, Python

Supervisor: Dr. Pradeep Kundu, Assistant Professor, KU Leuven

Auxiliary Classifier WGAN-GP for Time-Series Sensor Data Generation

• Built ACWGAN-GP with TCN discriminator to augment minority fault classes, reaching ~100% classification accuracy
• Evaluated synthetic data by comparing real and generated FFT spectra using PCC, Cosine Similarity, MMD, and KL Div

Conditional Latent Diffusion-GAN for CWT Generation and LiteFormer-based Classification

• Developed Conditional Latent Denoising Diffusion GAN (LDDGAN) with Supervised Contrastive Loss (SCL) for latent class separation (99.9% AE accuracy, 16× compression with EvoNorm-S0)
• Designed LiteFormer variants integrating Continuous Wavelet Convolution (CWC), Haar DWT, WDCNN-BiLSTM-Siamese Network, and CWMS-GAN-inspired architectures, achieving up to 99.18% fault classification accuracy

Tech Stack: PyTorch, Python

Supervisor: Dr. Kaushal Kishore, Principal Scientist, CSIR-CEERI

Exploring Unknown Indoor Environments with Drones Using a Multi-Critic DDPG Architecture

• Orchestrated a multi-critic DDPG control system for continuous pitch–yaw–roll drone navigation in indoor spaces
• Implemented a hybrid reward structure with artificial potential fields and precision replay into DDPG, improving navigation

Monocular Vision-Based UAV Navigation for Orchard Monitoring

• Engineered a UAV-based orchard monitoring system using YOLOv11 (Box mAP50: 95.5%, Mask mAP50: 96.5%)
• Programmed B-spline trajectory logic and implemented custom yaw-roll controller to minimize drift under wind

Tech Stack: PyTorch, Python, ROS Noetic, AirSim, NVIDIA Jetson Orin NX

CS F320: Foundations of Data Science
ME F218: Advanced Mechanics of Solids
ME F216: Materials Science and Engineering

• Assisted 100+ students in labs and tutorials, clarifying concepts and linking theory to practical applications
• Evaluated assignments and supported faculty in delivering high-impact teaching sessions

• Coordinated 10+ events and career sessions for 300+ students, facilitating technical exposure and alumni interaction.
• Managed production of 500+ merchandise items and led outreach, boosting student participation by 20%.

• Led a team of 25+ members to organize 4+ technical workshops with HVAC industry experts.
• Hosted 3 competitions and networking events, engaging 200+ students in HVAC innovation and awareness.

• Supervised 5 interdisciplinary robotics teams (30+ members) on Micromouse and Hexapod projects.
• Oversaw lab resources, conducted weekly reviews, and mentored 50+ students in hands-on technical skills.