• Setting up Industrial cameras for defect inspection and calibrating them to suit the use case. Image acquisition based on hardware trigger and data sorting for computer vision applications.• Design, execute, and easy to maintain reliable and bug-free code.• Train computer vision, and neural network models, optimize, and deploy. • Build resource-efficient and high latency AI models, to run efficiently on GPUs.• Continuously improve preprocessing and data pipeline to make predictions and run time faster.• Support in the installation process.

▪ Real time 6D pose estimation of surgical needle using only RGB images for treatment planning.▪ Research on using different data acquisition methods like point cloud, depth data and 2D images.▪ Data pre-processing to make the data useable for tracking small and precise objects like needles.▪ Synthetic realistic RGB image generation for deep learning models using Blender and BlenderProc.▪ Research on State-of-the-Art algorithms for pose estimation and real-time tracking.▪ Optimization and tuning of deep learning model.▪ Real-time streaming of data using gRPC to Jetson AGX for mathematical computations.▪ Simulations of the environment in Unity.▪ Real time testing and deployment of application on HoloLens2.

▪ Building Deep learning models for Container ship arrival prediction to destination ports.▪ Structuring data pipeline and pre-processing.▪ Informative dashboard for visualizing the ship tracking process on Power BI Maps.▪ Integrating pre-processing, training process, and visualizing predictions on Dashboard on the Microsoft Azure cloud platform.▪ Automating the complete cycle using ML-flow and scheduled refreshing the visualization.▪ Worked on API calls and web scraping for importing data from MarineTraffic and Fleetmon.

▪ Working on the vision system for TU Hamburg driverless formula student car.▪ Detecting multicolor cones using deep learning for path planning and simulation in the Gazebo simulator using bag files.▪ Object detection using TensorFlow Object Detection API for real-time obstacle detection on the track.▪ Building Object detection models for custom datasets to detect objects on the track.▪ Implementing algorithms in python and ROS (Robot Operating System).▪ Simulating the track environment inside ROS Rviz to analyze the algorithm robustness.

Project Arbeit: Anomaly detection in cyber-Physical systems using LSTM▪ use recurrent neural networks to analyse time series with typical industrial characteristics and forecast the data to detect the anomaly in the system.▪ Analyse the dynamic and static sensor data recorded at different time steps using LSTM.▪ analysing the forecasted data with and without anomaly and use the visualization and Gaussian techniques to interpret the results and find the abnormalities in the system.▪ Implementation in TensorFlow.

The ISIC challenge” Skin Lesion Analysis Towards Melanoma Detection”▪ Classification of skin lesion images using Machine Learning.▪ Tried different segmentation techniques and extract features like HOG, Harlick, Hu moments and classification using classical Machine learning techniques like SVM.▪ Later Applied deep Neural networks like CNN with feature extractors like efficientnetb4, resnet50, to achieve better sensitivity.▪ Built Multi Input-Output model in TensorFlow to build a model to have both metadata like age, gender, position of image taken and second input as images.▪ Model tuning and optimization using Augmentation, Hyperparameter Tuning and Transfer Learning.

Transcranial magnetic stimulation (TMS) is an effective medical procedure for diagnostic and therapeutic purposes. the procedure involves placing the magnetic coil to stimulate a specific cortex.▪ Developed a motion compensation system using ROS by tracking the head with a Kinect camera and moving a coil attached to a robot accordingly.▪ For this, used one of two depth cameras (Microsoft Kinect v2). Also, programmed to control a robot (UR3) to keep the coil at the target position relative to the head using 3d point cloud and ICP algorithm.▪ Additionally, calibrated the camera and the robot, in order to update the robot poses relative to the tracked head

▪ I was into Research and Development for 2-wheeler projects.