Graduate Student (2027 spring)
Backend Engineer, Machine Learning Engineer
Employers can request a warm introduction to this candidate.
Summary
Through my academic, research, and industry experiences, I have worked extensively with Python, C++, PyTorch, TensorFlow, LLMs, RAG, multi-agent systems, and machine learning infrastructure. In my current role as an AI Engineering Intern, I develop LLM-powered systems involving structured extraction, multi-agent workflows, tool calling, verification, and persistent state. I have also worked on efficient LLM serving through my graduate research at UC San Diego. My previous research experience has given me a strong foundation in designing experiments, developing scalable machine learning systems, and evaluating model performance. I have contributed to research in multimodal learning, algorithmic fairness, and language models, with publications in ACL and CHIL. In addition, I have developed backend AI applications and independent projects involving RAG pipelines, real-time voice systems, and automated LLM evaluation.
Experience
AI Engineering Intern
WorkCur Inc. Aug 2026 - present* Architected multi-step LLM chains and structured extraction flows to parse messy, unstructured resumes, engineering system prompts and typed schema validation with post-processing to eliminate parsing hallucinations. • Designed a dual-tier AI system introducing an autonomous multi-agent workflow (High-Effort mode) alongside a cost-efficient pipeline, wrapping deterministic verifiers into in-flight tools for real-time fact-checking and ATS alignment during generation. • Engineered a self-evolving partitioned state store over PostgreSQL us
Research Intern
University of Houston & INAOE May 2024 - Dec 2024* Designed a modular, scalable implementation of the HICCAP system (BERT, VGGish, I3D) using PyTorch, replacing tightly coupled monolithic components with well-defined, interoperable neural network modules. • Built a plugin-based fine-tuning framework with lifecycle interfaces across epoch, batch, backpropagation, and evaluation stages. Implemented six dynamic training strategies including a novel Dynamic Curriculum Learning method, raising multi-task macro F1 from 69.34 to 70.96 and binary F1 from 73.38 to 79.31 over the fixed-weight baseline to improve SOTA on comic mischief detection. • Unified separate binary and multi-task codebases into a single-pass training pipeline, enabling simultaneous training, evaluation, and testing of all five tasks and cutting redundant computation.
Software Developer assisting Dr Guanqun Cao
Michigan State University Oct 2023 - May 2024• Developed Python modules to select optimal hyperparameters for LassoNet classification and regression models. • Automated experiment execution on MSU’s HPC cluster using Linux shell scripts for job submission, result gathering, and post-processing.
Software Developer
Health Alliance Plan (HAP) Aug 2023 - Dec 2023Designed and developed backend services using Python Flask, OpenAI APIs, and LangChain to deliver a real-time conversational chatbot and a personalized recommendation system. • Designed the chatbot pipeline to retrieve and serve dynamic, real-time responses from HAP.org content, minimizing user navigation overhead. • Implemented the plan selector tool backend to process customer survey responses and apply AI-driven matching logic to recommend optimal health plans.
Research Intern
NSF – REU, Missouri University of Science and Technology May 2023 - Mar 2024• Researched algorithmic fairness in kidney allocation systems to analyze bias across sensitive attributes including race, gender, and age. • Implemented mathematical models in Python for Participant-Wise Learning and Fairness Feedback using logit-based discrete choice modeling and gradient descent optimization. • Developed and executed simulation experiments in Python to validate theoretical fairness findings across multiple group fairness notions.