Led the design & development of a vision-based LLM for robots. Initially, fine-tuned a foundation model with LoRA. Later, integrated it with a 'Point Transformer-2' architecture using modified geodesic multi-head self-attention for enhanced robot & object interaction boosting number of pick & place operations.

Developed an end-to-end deep learning model that regresses the generalised object representations as visual dense descriptors (embeddings) while additionally regressing object pose invariant 3D keypoints. The DNN (Deep Neural Network) was optimised using a self-supervised geometry aware loss functions, eliminating the need for manual annotated labels. The master thesis was awarded a '1.3' which is an excellent grade.

Implemented oriented bounding-box pruning & grouping for visual object pose registration using a DNN regressed feature space. The DNN feature space was optimized on the pixelwise distributional loss & agglomerative hierarchical clustering.

Integrated a robot physics engine from scratch, introducing hybrid motion force control to the NYU-Finger robot (a leg of a quadruped robot) with the in-house fabricated capacitive haptic sensor with robust real-time state estimation.

Commissioned 48 industrial robots, 2D/3D vision, machine tending & IoT systems as part of the design & site deployment team. The technical & commercial proposals pitched generated revenues more than the set yearly target.

Improved robot path planning by 60% fewer episode lengths using DDPG agent equipped with PER + HER + PENE methodologies using Tensorflow 2 & PyTorch. Furthermore, designed information maximization variation encoder (Info-VAE) for high data dimensional analysis & decomposition to its latent state, securing top-1 in the class.

Led student team of size 10 to participate in ABU-Robocon international event engineering 2 badminton playing robot. Furthermore, built a delta-style parallel robot for the bachelor's capstone project. Graduated with '8.99' CGPA which is equivalent to 'First Class with Distinction'.