Resume

Master of Science

Northwestern University

Machine Learning and Data Science

• Deep Learning
• Large Foundation Models
• Natural Language Processing
• Predictive Analytics
• Cloud Engineering

Bachelor of Science

University of California, Irvine

Data Science

• Machine Learning
• Probability and Statistics
• Algorithms
• Big Data Analytics
• Information Retrieval
• Database Management

Research Assistant - Computer Vision & Agricultural Robotics

US Department of Agriculture (USDA) Agricultural Research Service (ARS)

• Developing real-time computer vision system enabling autonomous weed detection and precision spray control on Farm-ng Amiga robotic platform
• Architecting multi-process system with IPC and threading to ensure low-latency integration of depth sensing, GPS, and camera feeds
• Deploying PyTorch models on NVIDIA Jetson Xavier via TensorRT FP16, achieving 8.6x inference speedup (718ms → 84ms)
• Engineering dual-camera GUI integrating weed detection and obstacle avoidance with automated safety-stop functionality

Data Science Consultant

Kavi Global

• Developed an LLM-powered lead qualification chatbot to route website visitors across 3 persona flows and drive consultation bookings—in production since February 2025
• Achieved 81% response accuracy across 180+ test cases by engineering GPT-4o prompts and dynamically leveraging website content as a RAG knowledge base
• Designed intent classification and prospect qualification workflow from greeting through service matching and case study recommendations; customized human handoff escalation for complex queries
• Recommended MS Copilot Studio over Dialogflow, Rasa, and Botpress for cost efficiency and Azure OpenAI integration

Data Science Engineer Intern

SS&C Intralinks

• Engineered a microservice-ready language identification pipeline for scanned PDFs using CLIP and PyTorch, achieving 98% accuracy across 14 languages (~0.3s/doc)—projected to save $300K/year on OCR costs
• Designed image preprocessing to select text-heavy pages, improving top-1 accuracy from 89% to 98% by filtering blank and image-heavy content before inference
• Benchmarked 30 vision-language models on AWS GPU instances using Hugging Face Transformers for accuracy, latency, and memory trade-offs
• Implemented adaptive multilingual prediction with configurable confidence thresholds for OCR and RAG pipelines

Full Stack Software Developer (Volunteer)

Irvine Canaan Christian Community Church

• Created a full-stack enrollment and attendance management system using Python, Flask, and MySQL for children's ministry, processing 80+ weekly check-ins/check-outs for 3+ years (since December 2022)
• Developed 8 data management dashboards and 15+ role-based pages serving 4 user roles with real-time validation
• Refactored REST APIs and reduced codebase by 3,600+ lines via SQLAlchemy optimization and DRY architecture
• Integrated barcode SDK for automated badge printing, enabling streamlined check-in workflow for staff and guardians

Data Analyst Intern

Shanghai Daiqian Information Technology Co., Ltd.

• Conducted consumer research for pre-launch product testing, performing survey analysis and building internal tooling
• Analyzed 190+ consumer surveys using R; findings: 9/10 satisfaction, 86% recommendation rate, 87% purchase intent
• Created 20+ visualizations using ggplot2 (satisfaction radar charts, demographic profiles, efficacy heatmaps, usage treemaps), adopted in stakeholder presentations and consumer testing reports
• Built internal trial management system on Alibaba Cloud with automated tester registration, feedback reminders, and sample tracking for product development workflows

paper2summary

• Developed a scientific paper summarization system by LoRA fine-tuning Llama-3.2-1B-Instruct on 20K arXiv papers, training only 0.07% of parameters (~850K) with 10K token context support (~28 hours on single RTX A6000)
• Achieved +51% ROUGE-2 and +37% ROUGE-3 improvement over base model on 6,440-sample test set

Dillard's Black Friday Return Prediction

• Built ML pipeline to predict Black Friday purchase vs. return outcomes, reducing return-related costs for Dillard’s
• Queried 160M+ POS records and applied SMOTE for class imbalance; trained K-means + Logistic Regression ensemble
• Achieved 78% purchase precision and 58% return recall with 227% projected ROI (~$590K)

Cum Laude

University of California, Irvine

Top 16%

Dean's Honor List

University of California, Irvine