KUMAR VISHAL

AI Architect | Principal Engineer | Deep Learning Specialist

 +91 77990 74798   kumarvishal.01@gmail.com
 linkedin.com/in/kumarvishal01   github.com/kumarvis

Summary

Seasoned technology leader with 15+ years in software design and development, specialising in Computer Vision, Deep Learning and Machine Learning. Proven record of delivering production-scale AI systems, securing patents and driving impact across semiconductor, healthcare, retail and automotive sectors. Seeking to spearhead architecture and innovation as an AI Architect / Principal Engineer.

Experience

KLA

• Architected Network for DefectWise AI wafer-defect-classification tool.
• Built a robust MLOps pipeline for rapid, repeatable deployments.

Barco

• Developed a predictive-maintenance platform for cinema projectors, analysing 900 GB of telemetry to forecast failures 5–7 days in advance.
• Created an AI ticket-triage system (LLama-3 prompts + BERT + LoRA-tuned Phi-3) that auto-tags issues and drafts resolutions with >96 % accuracy.

Entrupy

• Working on the solution for authenticating high value luxury items specially bags from the brands like Gucci, Louis Vuitton, Prada etc.
• Revamped mask-segmentation pipeline; Avg. Dice from 0.44 to 0.81 while eliminating class bleeding and contour errors.
• Redesigned key-point head, reducing number of poor-fit samples (error > 5 px) from 90 to 30.

GE Healthcare

• Worked on X-ray camera features to make the X-ray machines more smarter and efficeint to use..
• Patient-motion drift analytics—built an OpenVINO-optimised pipeline for the Definium X-ray platform that tracks movement in real time, averting ≈ 30 % of retakes (Product Demo).
• Automated collimator alignment—designed a Tradional Vision + DL model that predicts drift to ≤ 1° and drives closed-loop motor correction.

Cognizant Tech. Solution

• Delivered a two-stage Traffic-Light Recognition module—YOLOv2 for lamp-box detection plus an RGB-histogram CNN for state classification—achieving 0.94 F1 on the Bosch Small Traffic Lights dataset. (Traffic Light detection demo).
• Engineered a Road-Sign Recognition system for 40 sign types; clustered FC features to form 28 super-classes, then applied HOG + SVM sub-classification, boosting recall while maintaining real-time performance.

Samsung Electronics

• Co-developed a high-resolution (10 K × 15 K) image-processing library—runtime matched Matrox Imaging Library.

Sasken Communication

Red Brook

Education

IIIT Hyderabad — M.S. by Research, Jan 2013 – Apr 2015 | CGPA 8.25

AKGEC Ghaziabad — B.Tech in Computer Science, Aug 2002 – Jun 2006 | 62.3 %

Languages

English (Fluent) | Hindi (Native)