Arnav Gupta

Duke University Applied Machine Learning Lab

• Working on EEG-based brain-computer interface (BCI) systems, specifically P300 spellers designed for assistive communication in ALS patients.
• Research focuses on generalization challenges across users, analyzing trade-offs between data quality versus data quantity in training datasets.
• Implemented deep learning and classical ML approaches including EEGNet and SWLDA.
• Built full preprocessing pipelines for EEG data: signal filtering, epoch extraction, and noise or artifact handling.
• Generated feature-weight and model interpretability visualizations to better understand neural signal contributions.
• Contributing to a research paper (pending publication, NeurIPS dataset track involvement).

Big Fin Scientific

• Built DCS LinkStream 2.0, an Android-based system for ecological and environmental data collection.
• Integrated multiple hardware sensors into a unified data ingestion pipeline.
• Designed real-time syncing architecture for low-connectivity and offline-first environments.
• Improved reliability of field data collection across distributed teams and reduced manual data processing time by 75%.
• Supported 30+ global clients, including government and research organizations such as USGS and NOAA.
• Focused on data pipeline design, mobile systems engineering, and real-world deployment constraints.

4th Vector Technologies

• Developed computer vision pipelines for industrial defect detection in manufacturing.
• Used Python and OpenCV to preprocess image datasets, extract features, and train classification models.
• Built models across multiple defect types with high variability and achieved about 90% defect reduction.
• Delivered production-ready solutions used in real manufacturing environments.
• Contributed to $100K+ annual cost savings by preventing defects and recalls.
• Worked with clients such as Bridgestone and GSK.

• Focused on P300-based speller systems for assistive communication.
• Key research problem: improving cross-subject generalization in EEG-based ML models.
• Compared deep learning (EEGNet) and classical ML (SWLDA).
• Investigated dataset size versus diversity trade-offs and signal preprocessing impact.
• Built end-to-end ML pipelines for neural decoding.
• Generated interpretability visualizations for model understanding.
• Contributing to bigP3BCI dataset research (NeurIPS submission).

• Built ML models for cardiac arrhythmia detection using ECG data.
• Focused on feature engineering for time-series signals and robust preprocessing.
• Achieved 94% classification accuracy.
• Deployed model through a Flask-based application for real-time use.
• Presented at conferences and received 1st Place Catalyst Award.

• Designed an end-to-end defect detection pipeline.
• Used OpenCV and ML models for classification.
• Handled multi-class defect scenarios.
• Reduced defects by 90% in production systems.

• Built an Android-based data collection and syncing system.
• Integrated heterogeneous sensor inputs.
• Designed an offline-first architecture for field reliability.
• Used in real-world environmental deployments.

• Built a system to process and visualize large-scale SSD testing logs.
• Parsed 1,000+ JSON files into structured datasets.
• Developed a Flask dashboard for real-time analytics.
• Implemented anomaly detection for hardware testing workflows.

Duke University

B.S. Electrical & Computer Engineering and Computer Science

High School

High School Name (Placeholder)

Add your high school details, graduation year, awards, and notable coursework here.