Experience

Generative AI at Cornell

August 2026 – Present
Applied AI Researcher Ithaca, NY
  • Built a compaction method for coding agents with 93% instruction retention vs. 85% for full context at 1/3 the tokens
  • Beat 27 common compaction methods on LongMemEval at 98.7% accuracy & matched the best method on τ-bench at 8%

MotherDuck

May 2026 – August 2026
Software Engineering Intern New York, NY
  • Built the internal eval harness benchmarking AI agents on 40 user workflows via the MotherDuck CLI and MCP
  • Examined failed agent trajectories to write CLI guides for 3 features, taking task category accuracies from ~16% to 100%
  • Reduced agent token cost on the MotherDuck CLI up to 57% through A/B testing 5 CLI output optimizations
  • Designed & shipped CLI secret management commands, enabling AI agents to deploy workflows without seeing credentials

Vanasse Hangen Brustlin, Inc.

January 2026 – May 2026
Software Engineering Intern Remote/Raleigh, NC
  • Designed an agentic PDF-to-form pipeline for car crash reports, cutting analyst time by 66% over 10k+ crashes annually
  • Engineered an automated geocoder & RAG classification layer w/ project docs as context, achieving ~90% accuracy

Duke University

August 2024 – December 2024
Research Intern Durham, NC
  • Published first-author IEEE paper on CNN ensemble for Multiple Sclerosis diagnosis; 97.5% accuracy & 97.8% AUC-ROC
  • Fine-tuned VGG19/ResNet50 layers via depth-wise separable convolutions, reducing model size 14x for edge deployment
  • Injected Gaussian white noise alongside rotation, shear, and zoom augmentation to improve generalization by up to 7%

North Carolina State University

February 2024 – September 2024
Research Intern Raleigh, NC
  • Built a skin cancer classifier for edge devices with knowledge distillation, shrinking it 18x and lifting accuracy 19%
  • Trained a generative adversarial network to synthesize training data, improving overall performance score by 13%