My Experience
Jan. 2025 - Present
Nomad AI LLC
Software Engineer
• Contributed to backend itinerary-generation workflows serving 1,000+ monthly users, improving recommendation consistency based on 300+ internal QA evaluations. • Improved API throughput by 20% during load testing across hundreds of itinerary-generation requests, after profiling and optimizing data-flow logic. • Reduced average API response latency by 130 ms using lightweight caching and optimized request-handling pathways.
Mar. 2025 - Present
AIEA Lab @ UCSC
Undergraduate researcher
• Engineered a natural language–to–Prolog reasoning pipeline using GPT-4 and Python, supporting logic-based inference and improving structured query accuracy by 30% across a 500-sample evaluation set. • Built and optimized a hybrid RAG + Logic-LM framework that enhances retrieval precision by 35% and improves explainability scores by 20%, measured across 12 lab reasoning tasks.
Sep. 2025 - Present
Tech4Good Lab @ UCSC
web developer
• Co-developed Timely, a scheduling web app with HTML, TypeScript, Angular, and SCSS to automate coordination tasks and reduce manual scheduling time by 40% across 20+ workflow tests. • Improved UI responsiveness for 300 recurring student users, with internal usability studies showing 25% faster completion of common scheduling tasks.
Jun. 2025 - Sep. 2025
Kaiser Permanente
AI Software Engineer Intern
• Engineered a GPT-4o billing-summary generator evaluated on 150+ real patient billing statements, which reduced follow-up clarification questions by 25% across a pilot group of 40+ patients. • Developed an insurance plan-comparison assistant enabling users to complete comparison tasks 30% faster, measured across 25 plan-pair evaluations conducted during pilot testing. • Implemented a benefits Q&A chatbot that handled 200 support queries during controlled testing, reducing manual lookup time by 30 seconds per query for participating support staff.
May 2025 - Aug. 2025
Outamation
AI and Automation intern
• Built NLP + OCR pipelines (PyMuPDF, Tesseract OCR) that lowered manual document review time by 40%, measured across 1,200 internally labeled documents. • Constructed a LlamaIndex-based RAG prototype improving retrieval relevance by 18% on a 3,500-document internal knowledge base, benchmarked using 80+ representative test queries.