Open to opportunities

Sung Chun Lee

@sungchunlee

Message

Senior Applied Scientist building edge-to-cloud computer vision and multimodal LLM systems for safety.

United States

Message

What I'm looking for

I’m looking to build production-grade, safety-focused AI that scales from research to edge-to-cloud deployment—especially low-latency computer vision and multimodal LLM workflows—while driving rigorous MLOps, evaluation, and fast iteration.

I’m a Senior Applied Scientist specializing in computer vision and multimodal LLMs, with a focus on building scalable AI infrastructure that performs under real edge constraints. I take ideas from zero-to-one research through to agile, high-throughput production systems.

At Samsara, I architect and scale temporal computer vision pipelines for fleets of 1+ million edge devices, owning the end-to-end ML lifecycle. I led inward-facing drowsiness detection (resulting in US Patent US 12327417) and engineered outward-facing ADAS features like “Ran Red Light,” combining large-scale training, automated evaluation, and MLOps deployment. I also accelerate edge-to-cloud AI delivery using Cursor and multimodal LLM workflows to prototype, refactor, and ship production-grade safety algorithms.

Previously at Amazon, I built automated MLOps and evaluation pipelines by developing an RCA tool for the Easy-Check-Out (ECO) smart cart system. I spearheaded end-to-end research and production launch of specialized Just-Walk-Out (JWO) computer vision models for automated beverage dispensing, and engineered a 3D hand-fixture inference architecture that achieved a 20% reduction in over-charging errors.

Earlier at Alarm.com, I deployed real-time lightweight human tracking for a flagship doorbell camera under severe edge-compute and sub-second latency requirements. I also developed scalable object detection pipelines with robust MLOps, designed edge-AI facial detection/recognition with privacy-preserving local processing, and secured multiple US patents that translated foundational research into consumer IoT IP.

Experience

Work history, roles, and key accomplishments

Current

Senior Applied Scientist 2

Current

Samsara

Mar 2023 - Present (3 years 4 months)

Architected and scaled temporal computer vision pipelines for a fleet of 1+ million edge devices, owning the end-to-end ML lifecycle through low-latency inference on resource-constrained hardware. Designed inward-facing drowsiness detection and outward-facing ADAS features, deploying models via automated MLOps using AI-assisted development workflows.

Temporal Computer Vision Models System Architecture Drowsiness Detection Model Training

Applied Scientist

Amazon

Dec 2021 - Mar 2023 (1 year 3 months)

Built automated MLOps and evaluation pipelines by developing a root-cause-analysis tool for the Easy-Check-Out (ECO) smart cart system. Led research and production launch of Just-Walk-Out (JWO) computer vision models and engineered a 3D hand-fixture inference architecture to reduce over-charging errors.

MLOps Evaluation Pipelines Root Cause Analysis Computer Vision Product Recognition ML Networks Automated Beverage Dispensing Models Just Walk Out Systems 3D Inference Architecture Real Time Human Object Interaction Analysis

Senior Research Scientist

Alarm.com

Jan 2017 - Nov 2021 (4 years 10 months)

Architected and deployed real-time, lightweight human tracking models for a commercial doorbell camera, optimizing deep learning for severe edge constraints and sub-second latency. Developed scalable object detection pipelines with MLOps frameworks and designed edge-AI facial detection and recognition with privacy-preserving local processing.

Real Time Tracking Lightweight Deep Learning Detection Pipelines MLOps Edge AI Facial Detection Facial Recognition Identity Verification

Research Scientist

ObjectVideo Inc

Oct 2015 - Dec 2016 (1 year 2 months)

Architected advanced 3D scene modeling software (ImageExtruder) for precise geolocation and developed foundational 3D indoor modeling algorithms using dense point cloud data.

3D Modeling ImageExtruder Geolocation 3D Indoor Modeling Algorithms

Research Scientist

University of Southern California

Aug 2004 - Sep 2015 (11 years 1 month)

Led foundational computer vision research for video surveillance, designing real-time temporal anomaly detection and multi-object tracking systems for continuous outdoor video streams. Pioneered automated 3D spatial reconstruction systems, including algorithms for generating large-scale architectural models and performing robust camera calibration using multimodal inputs.

Video Surveillance Anomaly Detection Multi Object Tracking Multi Camera Calibration 3D Modeling