CV

Lead Engineer

Gordian | Remote

• Leading the research and development of Gordian Sense, a Computer Vision based cloud SaaS capable of analyzing store shelves and delivering insights to improve inventory management.
• Software Architect
  • Led the architectureal design and development of alexander - a modular framework for deploying computer vision pipelines in the cloud via configurable plugins for multiple deep learning models.
  • Key design feature include the ability to run plugins asyncrhonously via a Producer/Consumer execution paradigm, and an event-logger with appropriate log levels for critical errors, warnings and routine runtime info.
  • Defined template plugins for the engineering team to adopt and implement. This allowed for plug-and-play use of cutting edge deep learning models without breaking the pipeline.
• Engineering Team Lead
  • Established sound software practices by encouraging the engineering team to strive for full test coverage, clean code and detail-oriented PR etiquette.
  • Led and managed a team of computer vision, MLOps and backend engineers.
  • Brought in transparency and boosted pace of development by adopting scrum ceremonies - daily standups, task planning, reviews and retros.
  • Led regular code reviews to ensure maintainability and reproducibility
• Skills & Tools
  • Python, Bash, PyTorch, Docker, AWS {S3, Batch, EC2} Open3D, Numpy/Scipy, sk-learn
  • Git, Agile, Scrum, Unit Testing, CI/CD, GitHub Actions

Computer Vision Engineer II (L4)

Path Robotics | Columbus, OH

• Worked on Path's Adaptive Welding product - Multi-Pass Adaptive Fill
• Rapid prototyping and deployment of feature requests, unlocking welding patterns and stability improvements.
• Successfully led several RaaS (Robotics-as-a-Service) deployments of autonomous welding cells worth millions in revenue.
• Model Lifecycle Management - data engineering, data labeling, data science, MLOps, deployment
  • Conceptualized machine learning model to capture and generalize human welder preferences for multi-pass welding.
  • Handled data collection, hosting, data labeling, cleanup, and setup ETL (Extract/Transform/Load) pipeline.
  • model architecture selection, training, A/B testing, field trial and alpha deployment
  • Created a proof-of-concept for model serving and tracking with MLFlow and Databricks
  • Deployed model to cull search space for a uniform grid search algorithm. Translated to 60% execution time reduction for weld optimization.
• Skills & Tools
  • Python, C++, Bash, PyTorch, AWS CLI, EC2, S3, Databricks, Open3D, Shapely, Numpy/Scipy, sk-learn, PCL, MLFlow
  • Git, Agile, Scrum, Unit Testing, CI/CD

Computer Vision Engineer (L3)

Path Robotics | Columbus, OH

• Perception Team
  • Wrote a point cloud registration benchmark utilizing simple raycasting in Open3D.
  • Co-developed a 3D point cloud stitching algorithm using relative transforms from point-to-plane ICP to build a pose graph. Pruning false edges with uncertain alignment yielded true alignment between point clouds.
  • Assumed code ownership of the point cloud stitching ROS service. Handled field error reports and bugfixes.
• Adaptive Welding Team
  • Wrote an algorithm designed to find the closest interaction point for 2 near-orthogonal non-planar surfaces. (2D cross-section of a mesh).
  • Spearheaded a major refactor of Path's adaptive welding software stack. This transitioned the software from pre-release to true production-ready status, improving maintainability and stability. This kept the release stable in production until version 2 was developed.
• Skills & Tools
  • Python, C++, Bash, Open3D, Shapely, Numpy/Scipy, sk-learn, PCL
  • Git, Agile, Scrum, Unit Testing, CI/CD

Machine Learning Engineer

Streamn Inc | Cupertino, CA

• Rescued a prototype IPTV capturing backend with expert use of FFMPEG, Docker and cron management. Reduced server outages by 20%, dropping the recording failure rate from 40% to 7%.
• Worked on video-scene change detection with graph-based community detection.

Research Assistant

COSMOS Project | New York, NY

• Cloud Enhanced Open Software Defined Mobile Wireless Testbed for City-Scale Deployment (COSMOS) - part of the NSF PAWR program, partially funded by NSF award CNS 1827923.
• Advisor - Prof. Zoran Kostic
• Smart City Intersections
  • Led an analytical study† on real-time object tracking with YOLOv4, to define and measure the effect of three parameters against inference mean average precision (mAP) - scene complexity/object density, video resolution, and framerate. This work was published in proceedings of the 19th IEEE International Conference on Smart City, Dec 2021.
  • Explored accelerated Mask R-CNN inference with TensorRT and NVIDIA DeepStream. Performance differentiators were identified through CUDA profiling.

Research Associate

Spectrum Lab, Indian Institute of Science

• PI - Prof. Chandra Sekhar Seelamantula
• Medical Advisor - Dr. M. L. Murali Krishna
• Image Processing Project
  • Wrote an image processing library in Python to pre-process medical image data (retinal images, ultrasound images).
  • This work focused on multi-scale image processing (gaussian pyramids) and an image structure tensor derived from the Riesz Transform.
• Deep Learning Projects
  • Developed a segmentation system using U-Nets to identify features of pathological interest in fundus images (i.e. of the retina), such as vasculature, lipid deposits, optic disc etc.
  • Co-created a 3-stage Diabetic Macular Edema (DME) Severity prediction tool, utilizing U-Net segmentation to localize hard exudates and lipid deposits on the retina. The relative proximity and frequency of these exudates to the fovea determines the clinical severity of DME.
• Graphic Design Project
  • Created a cohesive design language using M1 Material Design principles for a pre-screening app for retinal image scans. This app was designed to be used by ophthalmologists and eye surgeons to screen medical images for pathologies using Spectrum Lab's research-backed detection algorithms.
  • This app was eventually demoed (with the same design) by Spectrum Lab at the G20 Science Summit in India (2023).

M.S.

Columbia University, Fu Foundation School of Engineering & Applied Science

• Major
  • Electrical Engineering
• Specialization
  • Data-Driven Analysis and Computation
• Projects
  • Real-time Object detection with YOLOv4
  • PCA with PyCUDA through Jacobi Method of SVD
  • Image Super-Resolution through Sparse representation of features from a learned dictionary
• Teaching
  • Seminar for ECBM E6040 Neural Networks & Deep Learning Research - A Tutorial on Adversarial Learning
  • Teaching Assistant for EECS E4750 Heterogenous Computing for Signal & Data Processing. (Course assistance, projects and grading).

BTech, Electrical and Electronics Engineering

Manipal Institute of Technology, Manipal

• Minor
  • Signal Processing and Control Theory
• Thesis: Application of Multi-scale Low-Rank Image Decomposition for Optic Disc Detection, under supervision of Prof. J. R. Harish Kumar.