Research

Research Interests

Key Areas: AI for Personalization & Recommendation Systems • Human-Centered AI • Multimodal Machine Learning • Interpretable AI • Behavioral Modeling & User Understanding • Spatio-Temporal Representation Learning • Large-Scale Data Systems • Privacy-Preserving ML

My research focuses on modeling human behavior and preferences using multimodal data from mobile and wearable devices, combining deep learning with human-centered approaches. I specialize in transforming complex time-series data into interpretable representations and building personalized systems that adapt to individual users. My work spans health applications, recommendation systems, and behavioral prediction at scale.

Research Experience

Research Assistant | Human AI-Technology Lab, University of Virginia

Aug 2020 – July 2025

• Designed a novel spatio-temporal representation framework for sequential user data, delivering 21% accuracy gain in one-shot inference—applicable to personalized content recommendation and user state prediction
• Created ranking framework for sparse multimodal data, achieving 13% recall improvement in predicting user state changes—directly transferable to engagement prediction and churn analysis
• Built personalized ML pipeline achieving 9x error reduction (RMSE 2.83 vs 25.7) through individual user modeling—demonstrating effectiveness of personalization strategies for large-scale systems

Research Assistant | VICTOR Lab, Virginia Tech, UVA

Aug 2018 – Aug 2020

• Developed patented autonomous sorting system with computer vision, reducing operational workload by 75%—demonstrating scalable automation for logistics and warehouse optimization
• Engineered path planning and computer vision solutions for autonomous systems in competitive robotics environment

Research Fellow | Wireless@VT Lab, Virginia Tech

Aug 2016 – Aug 2018

• Investigated reinforcement learning (multi-armed bandit) algorithms for resource optimization in DARPA Spectrum Challenge—applicable to A/B testing, explore-exploit problems, and dynamic recommendation systems
• Implemented comprehensive testing and evaluation frameworks for algorithmic performance in competitive environments

Featured Research Contributions

Visual Encoding of Time Series: A Voronoi-Based Framework for Behavioral Pattern Discovery

Human AI Technology Lab | May 2022 – Ongoing

Topics: User Behavior Modeling, Representation Learning, Contrastive Learning, Pattern Recognition

Developed a novel approach to transform raw sequential user data into interpretable visual representations, enabling pattern discovery across diverse applications—from content consumption patterns to user engagement trends.

Key Innovations:

• Computational geometry-based encoding that preserves temporal and spatial relationships in high-dimensional behavioral data
• Contrastive learning framework for discovering latent user segments without explicit labels
• Color-based visualization strategy that reveals anomalies and behavioral shifts across user populations
• Transferable representation applicable to recommendation systems, user modeling, and engagement prediction

Applications:

• User segmentation and cohort discovery
• Personalized recommendation systems
• Engagement pattern recognition
• Anomaly detection in user behavior

Status: Paper submitted to ACM HEALTHCARE 2025

FLMS: A Hybrid Framework for Ranking Predictions from Longitudinal Multimodal Data

Human AI Technology Lab | Aug 2021 – Dec 2023

Topics: Multimodal Data Fusion, Personalized Modeling, Ranking Systems, Cold-Start Problems

Developed FLMS, a hybrid framework that addresses the cold-start problem in personalized systems by intelligently balancing population-level and individual-level models. The framework ranks predictions from sparse, longitudinal data—critical for applications with limited user history.

Key Contributions:

• Tensor-based aggregation for multimodal data fusion across diverse signal types (text, images, sensor data, engagement metrics)
• Dynamic weighting strategy that adapts between global and personalized models based on data availability
• Interpretable ranking mechanism that explains prediction confidence—crucial for trust in recommendation systems
• Validated on real-world sparse data demonstrating robustness in low-data regimes

Industry Applications:

• New user onboarding and cold-start recommendation
• Multi-signal ranking for content discovery
• Personalized notification systems
• User state prediction for proactive intervention

Publication: JMIR-AI

MoodRing: Predicting User State Changes from Multimodal Passive Sensing

Human AI Technology Lab | Aug 2020 – Jun 2022

Topics: Behavioral Analytics, Predictive Modeling, Mobile Sensing, User Engagement

Built an end-to-end system for predicting user state changes using multimodal passive data streams. Demonstrated how continuous behavioral signals can predict future engagement patterns and user needs.

Technical Achievements:

• Feature engineering pipeline extracting interpretable signals from noisy, high-dimensional time series
• Temporal modeling capturing both short-term fluctuations and long-term trends in user behavior
• Feature importance analysis identifying key behavioral indicators—directly applicable to engagement metrics and user analytics
• Real-world deployment validating system performance at scale

Transferable Skills:

• Time series forecasting for user engagement
• Multi-signal feature extraction and selection
• Predictive analytics for proactive recommendations
• Privacy-preserving behavioral modeling

Publication: JMIR Formative Research

Autonomous Smart Manufacturing: Blockchain-AI Integration for Decentralized Systems

External Collaboration | Aug 2020 – Aug 2021

Topics: Distributed Systems, Federated Learning, AI Optimization, Secure Data Sharing

Designed a decentralized framework integrating AI with blockchain for autonomous optimization across distributed entities. Addressed key challenges in multi-party collaboration: data privacy, trust, and efficient optimization.

Key Innovations:

• Federated learning architecture enabling collaborative model training without centralized data aggregation
• Blockchain-based trust layer ensuring transparent, verifiable transactions across parties
• ML optimization algorithms for resource allocation, predictive maintenance, and decision-making
• Privacy-preserving computation maintaining data security across organizational boundaries

Relevant to:

• Federated recommendation systems
• Privacy-preserving ML for user data
• Distributed optimization at scale
• Multi-stakeholder AI systems

Publication: IISE

Core Technical Competencies

Machine Learning & AI

• Deep Learning (CNNs, RNNs, Transformers, Contrastive Learning)
• Recommender Systems (Collaborative Filtering, Content-Based, Hybrid Methods)
• Ranking & Personalization Algorithms
• Reinforcement Learning (Multi-Armed Bandits, Exploration-Exploitation)
• Time Series Forecasting & Anomaly Detection

Data Science & Engineering

• Large-Scale Data Processing & Feature Engineering
• Multimodal Data Fusion (Text, Image, Sensor, Behavioral)
• A/B Testing & Experimentation
• Causal Inference & Statistical Modeling
• Model Interpretability & Explainable AI

Systems & Deployment

• ML Pipeline Development & Production Systems
• Privacy-Preserving ML & Federated Learning
• Real-time Prediction Systems
• Computer Vision & Autonomous Systems

Publications & Patents

• ACM HEALTHCARE 2025 (Submitted) - Visual Encoding Framework for Behavioral Discovery
• JMIR-AI - FLMS: Hybrid Ranking Framework for Sparse Multimodal Data
• JMIR Formative Research - Predictive System for User State Changes
• IISE - Blockchain-AI Integration for Decentralized Optimization
• Patent - Autonomous Sorting System with Computer Vision

Research Impact

My work demonstrates end-to-end capabilities from research to production:

• Personalization at scale: Built systems handling sparse, multimodal data for individual-level predictions
• Novel representations: Created interpretable frameworks that bridge the gap between raw data and actionable insights
• Real-world validation: Deployed systems used by real users, not just academic benchmarks
• Cross-domain expertise: Applied similar methodologies across health, robotics, and distributed systems

I’m passionate about building AI systems that understand and adapt to individual users while maintaining privacy, interpretability, and trust.

Last Updated: September 2025