Software engineer and published researcher specializing in enterprise-scale AI, haptic robotics, and cybersecurity. Published for work on preemptive error correction in motor learning. Founded DefenX Security Solutions. Currently building agentic AI at FedEx Dataworks, optimizing LLM serving with NVIDIA, and evaluating LLM bias at UMD's Urban Computing Lab. Core contributor to Apache Kafka 4.0.
Experience
Memphis, TN (Remote)
- Implemented critical data migration pipelines processing 80M+ daily queries across global operations.
- Developed infrastructure optimization tools and agentic AI architectures for Customer Data Platforms.
- Engineered synthetic data libraries automating PII obfuscation while preserving referential integrity across complex multi-table schemas.
- Designed natural language-to-SQL workflows using LangGraph and Azure AI Foundry.
College Park, MD
- Researching LLM serving optimization with SGLang & vLLM inference frameworks.
- Exploring high-throughput inference techniques for production-scale deployments.
College Park, MD
- Co-authored FIXical I/O — a magnetic hand exoskeleton enabling preemptive error correction for finger-based motor sequence learning.
- Designed real-time error sensing combining IMU motion tracking with electromagnet-based actuation for haptic feedback during piano learning and rehabilitation.
- Engineered haptic optimization reducing tuning time from 55 seconds to 0.3 seconds.
College Park, MD
- Analyzing geospatial data from 15 million mobile devices to identify and mitigate AI prediction gaps.
- Evaluating LLM bias in geospatial contexts.
College Park, MD
- Founded DefenX — an AI-powered cybersecurity startup building next-generation threat detection and response systems.
- Raised $250K in pre-seed funding from institutional and angel investors.
- Leading product development, GTM strategy, and engineering for enterprise-grade security solutions.
College Park, MD
- Led 10+ engineers on nonprofit software development projects.
- Managed sprint planning, code reviews, and delivery timelines.
Open Source
- Implemented KIP-1087 for Apache Kafka 4.0 — contributed to the core distributed streaming platform.
Redmond, WA
- Built NLP data pipelines for the World Wildlife Fund.
Los Angeles, CA
- Built a full-stack React + Redux application displaying key metrics for city planners and policymakers.
- Aided in crime mitigation, homelessness reduction, and creating a safer, more equitable city.
- Met weekly with volunteer developers to discuss feature implementation and bug fixes.
Los Gatos, CA
- Worked on airline management system development using Apache Hadoop and Spark.
- Created ETL pipelines with Spark by standardizing and aggregating large datasets.
Research
FIXical I/O: Preemptive Error Correction for Motor Learning
DOI ↗UMD Embodied Dynamics Lab — Published
Co-authored a full paper on FIXical I/O — a magnetic hand exoskeleton combining real-time motion sensing with electromagnet-based actuation for Preemptive Error Correction in finger-based motor sequence learning. Achieves higher learning performance, error awareness, and confidence.
LLM Bias Evaluation in Geospatial Contexts
UMIACS / Urban Computing Lab
Analyzing geospatial data from 15 million mobile devices to identify and mitigate AI prediction gaps. Investigating systematic biases in LLMs when applied to location-based tasks and urban computing scenarios.
Transformer-based Synthetic Data Generation
FedEx Dataworks
Engineering synthetic data libraries that automate PII obfuscation while preserving referential integrity across complex multi-table schemas. Enabling privacy-compliant data generation for enterprise-scale testing and analytics.
Agentic AI Architectures for Enterprise
FedEx Dataworks
Designing agentic architectures for Customer Data Platforms using LangGraph and Azure AI Foundry. Building natural language-to-SQL workflows that enable non-technical stakeholders to query complex data systems.
LLM Serving Optimization
NVIDIA × UMD
Researching high-throughput inference techniques with SGLang and vLLM for production-scale LLM deployments. Exploring novel approaches to maximize tokens-per-second while maintaining response quality.
Publications
FIXical I/O: Exploring the Effects of Real-time Error Sensing and Physical Intervention on Finger-based Motor Sequence Learning
ACM ↗Kyungyeon Lee, Jai Vaichalkar, Arnav Dadarya, Wooje Chang, Atsushi Kikumoto, Jun Nishida
ACM · Barcelona, Spain
Introduced FIXical I/O, a magnetic hand exoskeleton combining real-time finger motion sensing with electromagnet-based actuation to enable Preemptive Error Correction — sensing emerging errors and physically nudging fingers away before mistakes occur. Demonstrated significantly higher learning performance, error awareness, and confidence in piano learning and rehabilitation tasks.
ImageBreak: A Framework for Testing AI Model Safety
GitHub ↗Arnav Dadarya, Anushk Pokharna · CMSC396H
A comprehensive framework for evaluating AI model safety and content moderation systems — systematically testing adversarial prompts, identifying vulnerabilities in content filtering, and providing quantitative safety assessments across multiple providers.
Education
University of Maryland
B.S. Computer Science · Minor: Statistics
El Camino Real Charter High School
High School Diploma
Achievements & Honors
- Published researcher — top-tier ACM conference on Human-Computer Interaction (17 pages, peer-reviewed)
- Founded DefenX Security Solutions — raised $250K in pre-seed funding for AI-powered cybersecurity
- Implemented KIP-1087 for Apache Kafka 4.0 (core open-source contributor)
- Built data-migration pipelines processing 80M+ daily queries at FedEx Dataworks
- Engineered haptic optimization reducing tuning time from 55 s → 0.3 s
- Technical Lead at Hack4Impact UMD — led 10+ engineers on nonprofit software
- Microsoft Hack4Good — NLP data pipelines for World Wildlife Fund
- Dean's List & CS Departmental Honors
- Technical Excellence and Mastery Award