Kshama Nitin Shah

Computer vision research is advancing at a remarkable pace, with new models for perception and multimodal understanding emerging almost weekly. Deploying these models in the real world is a different challenge entirely: production systems must run on constrained embedded hardware, in unpredictable environments, where reliability, latency, and safety matter as much as benchmark accuracy. In this talk, I will share lessons from translating computer vision research into production perception systems for commercial vehicles, including real-time multi-camera pipelines, detection and segmentation on embedded GPUs, and calibration-aware model design. I will cover the deployment pipeline end to end: data collection and curation, model design under compute constraints, optimization and quantization, on-vehicle integration, and continuous monitoring. I will also discuss what breaks between the lab and the road, from sensor drift to lighting variation, and how to build robustness without continuous retraining. Attendees will leave with a practical framework for evaluating new research, understanding what productionization actually requires, and building vision systems that are both state of the art and operationally dependable.

Biography:

Kshama Shah is an AI Research Engineer working on computer vision and multimodal AI, with experience spanning industry and academic research. She holds an M.S. in Electrical and Computer Engineering from the University of Michigan, where she worked on vision-language research, and currently builds real-time perception systems for automotive applications.

Marian Newsome

The work of AI governance, deciding what a system is permitted to do, proving it did only that, and catching it when it does not, is not new. It has been running in industrial control rooms for decades. This talk discusses a career path from gas field SCADA through enterprise audit, compliance, product, and security resilience, arriving at AI governance before the field had a name.

For engineers wondering whether their operational or compliance experience is a detour from a technical career, this talk argues the opposite: instrumentation and control backgrounds are now among the most transferable credentials in AI.

Biography:

Marian Newsome is the founder of Ethical Tech Matters, an independent AI governance advisory and consulting firm based in Fort Worth, Texas. She is one of the first 100 IEEE CertifAIEd Lead Assessors globally, following 18+ years at Dell Technologies across audit, compliance, product, and security resilience. She advises boards and executives in defense, rail, energy, healthcare, and critical infrastructure on AI governance.