Kanishk Navale

Built the company's AI framework to power their entry into AI-driven vulnerability management. Designed production ML pipelines for rapid LLM fine-tuning and deployment, delivering NLP models with 96% accuracy in vulnerability detection script generation. Spearheaded the design and implementation of a multi-agent AI orchestration platform. The system delivers context-aware mitigation recommendations, downtime impact classification, and intelligent remediation prioritisation used by all product customers.

Led abstraction & development of the company's specialized large language model product designed for robot interaction. Fine-tuned foundation models using LoRA methodology blended with geodesic multi-head self-attention for robot & visual object interaction. Engineered vision transformer extracting feature pyramid using PyTorch-Lightning framework for depth image completion with robust object mask estimation to infer valid depth images of an object and bring in geometrical aware boundary depth completion.

Graduated with distinction (grade 1.3/1.0) and published an IEEE paper. Developed a self-supervised visual representation model that strengthened robotic grasp success from 70% to 90% by designing an end-to-end network with geometric constraints, eliminating the need for manual data labelling. Accelerated model training by 200% and removed dependency on unreliable depth cameras by designing an RGB-only 3D keypoint regression architecture, enabling more robust pose estimation for production deployment.

Contributed pixel-wise distributional loss to optimize the deep learning network. The implemented loss function led the DNN to yield better uni-modal predictions visually by inspecting the heatmap of tracked keypoint over the existing loss function. Formulated an optimization process to regress geometrically consistent visual keypoints on the dense feature space by engineering self-supervised losses, eliminating manual keypoint selection to create oriented bounding boxes.

Developed a haptics-based robot physics engine as a research assistant. Collaborated with the internal electronics engineering team of size 4 to modularize capacitive-based haptic sensors enabling direct integration with the robot core operating system. Installed 'Hybrid Controller' in the robot, improving sensing by 55% by incorporating probabilistic state estimation for the haptic sensor.

Critiqued 3 student industrial automation & robotics research teams simultaneously. Cofounded an internal group acting as a consulting firm for local industries & research teams to fabricate economical automation & robot controllers.

Commissioned 48 industrial robots in welding, 2D & 3D vision, machine tending, IoT with predictive maintenance as part of the design & site team. The technical & commercial proposals generated revenues more than the set yearly revenue targets.

Improved robot path planning by 60% fewer episode lengths using DDPG agent with prioritized & hindsight experience replay with parametric exploration noise using Tensorflow 2 & PyTorch. Implemented transformers-based speech & language learning model for classification with model optimizations, earning top-2 position in the class. Designed information maximization variation encoder (Info-VAE) for high data dimensional decomposition to its latent state, securing top-1 the class.

Led student team of size 10 to participate in ABU-Robocon international event engineering two badminton-playing robots. Built a delta-style parallel robot for the bachelor's capstone project. Designed mechanical robot system components using the SolidWorks CAD tool & inspected the manufactured parts based on the GD&T standards. Furthermore, based on control design, custom PCBs were fabricated using the EAGLE tool, and a PLC-based motion system was deployed using Beckhoff TwinCAT.