Debjit Ghosh

Designing and building autonomous surgical robotic systems driven by image guidance and AI.

Autonomous Ground Vehicle for Agricultural Applications (under NDA) - Designed and engineered autonomy software for fleet of custom agricultural utility vehicles - Conducted extensive research while developing state estimation algorithms used for localizing the robot under constrains of its environment - Implemented sensor fusion for accurate odometry combining GPS, 9DOF-IMU, LiDAR and wheel encoders using Kalman Filters - Implemented Real-Time-Kinematics (RTK) for increased accuracy of localization in the presence of satellite connectivity - Implemented autonomous geospatial navigation capability to provided satellite coordinates - Developed SCADA for deploying multiple robots interfacing with hardware controls over PLC - Developed capability for remote deployment of autonomous vehicles - Collaborated with in-house software-engineers - Collaborated with mechanical engineering team for developing custom sensor mounts - Project management of full life cycle of development - Providing adequate engineering documentation, code revision history and deployable software - Communicating progress regularly to key stake holders - Balancing research and scientific engineering development goals while demonstrating progress

Autonomous Robot for Warehousing Applications using Real-Time Appearance Based (RTab) SLAM - Developed warehousing robot application using a 4WD custom robot capable of autonomously navigating to docking stations for carrying out pick and place operations - Developed custom URDF to interface with ros_control - Implemented SLAM using RGB-D and encoder feedback for mapping a warehouse for localizing during navigation - Implemented planning algorithms using ROS NavFn and TrajectoryPlanner - Further details subject to NDA

Contributed to developing the core software and scalable infrastructure for the world's largest autonomous 3D printer used for manufacturing both stages of rocket fuel tanks involving real-time closed loop control systems for streaming, processing and viewing live feed of the underlying non-deterministic welding process as well as development of hardware based real-time data acquisition systems.

• Teaching python, applied statistics and inferencing, visualizations, data engineering at Thinkful • Model building, feature selection, supervised and unsupervised learning algorithms, testing and evaluation of models • Undertaking multiple capstone projects and performing student evaluations • Data science toolkit: jupyter, pandas, numpy, sklearn, seaborn

• Used unsupervised machine learning and collaborative filtering for generating real-time personalized recommendations from millions of videos prototyped using R and engineered using Java • As lead data scientist, prototype, design, architect and built a big data platform for deploying the recommendations engine across US/EU with a content right management system • Real time data ingestion of user data and video content using Kafka and Spark Streaming, and storage using Cassandra • Cluster computation and storage in the AWS cloud using custom cookbooks for Spark-Cassandra nodes • Project management using agile principles across teams in web development, search, authentication involving management of project deadlines, triaging bugs, resource allocations, documentation using JIRA, conducting daily scrum for a large software engineering team • Gather requirements and designing the algorithm for Popular and Trending Videos • Languages and platforms used: R, Java, AWS, Spark, Spark Streaming, Kafka, Cassandra, HDFS

• Aggregating 15 years of time-series data across Cerner and ISM SQL Databases with over 20,000 ICU patients recorded over 25 years at Children’s Hospital Los Angeles (CHLA) • Preparing Dataset 2.0 which served for machine learning research and allowed for growth of team under Dr. Randall Wetzel’s mentorship requiring extensive cleaning and complex aggregations using C++ • Designed and implemented a HIPAA compliant de-identification algorithm that facilitated collaboration across UCR and USC • Normalizing age-dependancy in the pediatric ICU population using polynomial regression models of physiologic data • Generating features from non-uniformly sampled, multivariate, time-series data and applying unsupervised machine learning algorithms in order to discover medically meaningful clusters which further define phenotypes • Determined significant medically relevant information from PICU electronic medical record data which included prognosis, diagnosis, and therapy in an unsupervised approach • Published in the Journal of Pediatric Critical Care Medicine (PMID:29727354)

• Research project under the mentorship of experts in Neonatal Cardiology: Dr. Istvan Seri, Dr. Bijan Siassi et. al. for designing, building and testing a simulation of neonatal echocardiography for training physicians from across the globe at the Keck School of Medicine • Highly accurate 6DOF sensors based on electromagnetic transmitters and trackers with a C++ API • Visualization Toolkit (VTK) for 3D image processing and visualizations for slicing cardiac volumes acquired using echocardiography based on sensor movements over mannequin • Core engine written in C++ ensuring real-time performance, user interface designed using C# and software built, debugged and tested using Visual Studio • Complex slicing algorithm for visualizing views of the cardiac system by fixing at the apex of the cone of the DICOM echocardiogram volume and rotating across the image based on sensor readings • End to end software that was shippable and installable on any computer with a powerful graphics card. • Presented at 3 consecutive Academic Pediatric Association (APA) • Patented under Virtual Neonatal Echocardiographic Training System [Publication number: 20170337846]

• Participated as team member in the 6th mission of the AUVSI International Aerial Robotics Competition • SSH communications over ROS topics between the Arduino board and the Pico ITX • Serial programming of the Arduino motherboard • SLAM for predicting the position of the robot as accurately as possible based on the knowledge of its surrounding environment. Used Kinect, 2D video camera • Control system comprising on-board as well as remote processing for sending positional corrections based on accelerometer, gyroscopes and PID controllers, to the quadrotor

Expensive arithmetic operations of analyzing raw data ported to GPGPU (NVidia GeForce 9400 GT) using CUDA. Involved heavy usage of Native C overlaid with C#. Implemented Non-Destructive-Technology data manipulation and analysis algorithms. Developed UI for John Deere’s Drum Knife Seem Inspection Project (C#). Integrated core software, Kovid, with UI specifically for John Deere’s project. Prepared test cases for final software using MBUnit. Provided remote support for application deployed in John Deere. Implemented 3D visualization of the materials in VTK. Played a major role in designing the software for being deployed on the Web using PHP and Java 3D. Tested performance and memory consumption of core software using ANT Profiler. Developed various other components for the core software, Kovid on the .NET framework. Trained and mentored recruits.