Vinay Ch

I build behavioral modeling pipelines for machine learning operation kernels, including GEMM operations with different precision levels and activation functions, all tuned to the specific architecture and tested for model-level accuracy. I also work on improving the CARF flow to better handle ONNX graph manipulation and data extraction. These efforts help optimize our neural processing units, making inference faster and more efficient.

I worked on the Unified Knowledge Platform (UKP), an AI-powered document search system for factories that handles PDFs, Office documents, images with OCR, and Intel's custom formats, all while meeting strict security requirements. I also contributed to custom neural network architectures using auto-encoders for learning patterns and detecting anomalies in wafer images, which run on Intel's manufacturing cloud through the WISTA system to process huge amounts of fabrication data in real-time. I worked on several proof-of-concept systems using computer vision, OCR, image stitching, and visual search to find defects in manufacturing, which led to a production system for detecting laser damage with automatic notifications, plus a codeless object detection tool that engineers can use with simple labeling. I also had two abstracts accepted at IMEC 2022, an internal manufacturing conference with only a 1% acceptance rate, and contributed to the Intelidays holiday management portal for about 6,500 factory employees, replacing their old spreadsheet system.

I worked on designing a custom loss function and contributed to an end-to-end pipeline using siamese CNNs to estimate relative pose between voxel grids, which are a much more memory-efficient way to represent point clouds (about 90% smaller). This work was published at ICRA 2019 in the SLAM benchmarking workshop: "End-to-End Relative Pose Estimation of Point Clouds and Voxel Grids" (L. Rodriguez, V. Chandragiri, D. Pena, and D. Moloney) at Benchmarking of SLAM Algorithms for Robotics and VR/AR, ICRA 2019. I also worked on a low-power object detection system that can recognize 200 different object classes for the LPIRC 2019 challenge, with all the preprocessing and inference running on a Movidius Myriad-X chip. Additionally, I contributed to a complete low-power IP camera system that captures images via MIPI, compresses them with JPEG, runs neural network inference using OpenVINO, and sends results over HTTP, all on a single Movidius Myriad-X device.

I managed nine Master's and PhD interns over three years using agile methods, handling day-to-day operations and making sure projects got done end-to-end. I worked with teams across the US, China, and Israel to find new opportunities and improve how we work together internationally. I also ran the entire hiring process for interns and full-time hires for 2.5 years, working with universities in Ireland and Europe to find, interview, and hire candidates. On top of that, I organized a weekly Show & Tell learning session for four years and set up ergonomic assessments for the team over three years to keep everyone safe and healthy.

I worked on creating and running a program that found and brought in about 25 tech start-ups from across Europe to work with Intel Movidius. I coached these entrepreneurs on how to use Movidius software and hardware effectively, helping them build their products faster and get ready for market. I kept track of all the start-ups' progress, making sure they hit their milestones and got the support they needed, while also connecting them with the right people and resources.

I mentored over 1,000 students worldwide through Udacity and Thinkful programs, focusing on helping them actually use what they learn to solve real problems. I checked in with students weekly, had regular calls, and tracked their progress across courses in Data Science, Software Engineering, Computer Vision, Machine Learning, and Project Management. Many of my students successfully switched careers, especially during 2020-2021, and landed remote jobs in Data Science and Analytics.

I led a team of four people who managed 60 volunteers to put on Techniche 2014, a major tech festival at IIT Guwahati, organizing nine workshops, seven lecture series, and twelve competitions. The next year, I mentored a 65-person team to run the festival smoothly, helping everyone work together effectively and develop their leadership skills.