cv

Machine Learning Engineer Intern - LLM Post-training & ML System

Venera AI

Developed and deployed a scalable and efficient LLM post-training and ML system to improve the accuracy and efficiency of the model.

• Fine-tuned (SFT and distillation) Qwen3 to compress knowledge using QLoRA + DeepSpeed ZeRO-3
• Engineered a high-throughput inference for LLM; support prefix/KV caching, continuous batching.
• Deployed LLM on TPU v5 (2x4 pod, 8 chips), establishing a cost-efficient alternative to GPUs.
• Rebuilt data pipeline Spark with Ray Data, managed via Airflow, 3x throughput to 1M+ tok/min.
• Consolidated CI/CD (GitHub Actions + Terraform + Helm) and monitoring (Grafana + Prometheus).

Machine Learning Engineer Intern - AI Agent

Adobe

Developed and deployed a scalable and efficient AI agent to improve the accuracy and efficiency of the model.

• Developed Voice Agent features combining planning, speech recognition, and emotion-aware TTS.
• Built AI-agent using LangChain, LangGraph, and MCP to integrate into Adobe multi-agent system.
• Deployed production models via vLLM, Ray Serve, FastAPI, integrated with Kubernetes and ArgoCD.

Machine Learning Engineer Intern - Recommendation System

VCCorp Corporation

Developed and deployed a scalable and efficient recommendation system to improve the accuracy and efficiency of the model.

• Developed a scalable and efficient recommendation system to improve the accuracy and efficiency of the model.
• Developed a scalable and efficient recommendation system to improve the accuracy and efficiency of the model.

Master of Science

University of Maryland, College Park

Computer Science

• Machine Learning

Bachelor of Science

Gettysburg College

Computer Science

• Phi Beta Kappa Society member
• David Wills Scholarship recipient

ICPC Participant

ICPC

ICPC is a competitive programming contest for university students.

Predicting Perceived Music Emotions with Respect to Instrument Combinations

AAAI

Music Emotion Recognition has attracted a lot of academic research work in recent years because it has a wide range of applications, including song recommendation and music visualization. As music is a way for humans to express emotion, there is a need for a machine to automatically infer the perceived emotion of pieces of music. In this paper, we compare the accuracy difference between music emotion recognition models given music pieces as a whole versus music pieces separated by instruments. To compare the models’ emotion predictions, which are distributions over valence and arousal values, we provide a metric that compares two distribution curves. Using this metric, we provide empirical evidence that training Random Forest and Convolution Recurrent Neural Network with mixed instrumental music data conveys a better understanding of emotion than training the same models with music that are separated into each instrumental source.