Work
Experience and selected projects.
Experience
DeepTek.ai · ML Engineer
Shipped production computer vision systems for chest X-ray analysis and supported clinical and regulatory validation across global deployments.
- Developed PyTorch and TensorFlow computer vision pipelines for chest X-ray analysis, achieving over 90% AUROC for pleural effusion detection in deployed FDA approved clinical workflows.
- Architected an interactive experimentation platform with automated inference, MLflow based tracking, and metric computation, reducing model evaluation turnaround from several hours to under 5 minutes.
- Led regulatory validation across US FDA, Thai FDA, HSA, and CE approvals by coordinating an MRMC study with 24 radiologists to demonstrate AI system efficacy.
- Maintained a MongoDB annotation database for over 1.8M chest X-ray studies and led CVAT migration, ensuring data consistency and infrastructure reliability across clinical datasets.
Anheuser-Busch InBev · Automation Intern
Built analytics infrastructure and dashboards for employee work-pattern analysis, from raw task mining data through SQL storage to executive-facing Power BI reporting.
- Designed a Power BI dashboard on employee work patterns and built SQL database infrastructure to support recurring analytics workflows.
- Performed data wrangling on Task Mining API data with over 1M weekly records using Azure Data Factory for analytics pipelines.
iQGateway · Data Science Intern
Developed model diagnostic visualizations for AutoML pipelines, improving how teams inspect model behavior during training and selection.
- Built diagnostic visualizations for AutoML model evaluation, surfacing training behavior and model comparison signals to speed up debugging.
Prime Focus Technologies · Machine Learning Intern
Worked on automated audio processing and deep learning model evaluation for media production pipelines.
- Wrote Python scripts for automated audio processing and feature extraction to support production model evaluation workflows.
- Evaluated deep learning models for audio source separation, comparing candidates on separation quality and runtime to guide production model selection.
