Anthony Han

About Me

Profile photo of Anthony Han

Anthony Han

Ex-AI Engineer Intern @ IBM | Mechanical Energy Engineering @ CAU

Atlanta, GA

I'm an undergraduate student at Chung-Ang University, majoring in Mechanical Energy Engineering. My interests mainly lie in Data Science, Developing AI agents, and in Recommendation Systems. I'm also finding ways to integrate those areas with my mechanical/energy engineering background.

Previously, I've worked as an AI Engineer Intern at IBM, developed Multi-Scale AI agents and AI-powered custom IT management dashboards (ITAM) for clients detected from IBM's observability solution (IBM Instana).

Education

Chung-Ang University

B.E. · Mechanical Energy Engineering

Mar. 2021 - Feb. 2027 (Expected)
Degree
Bachelor of Engineering (B.E.)
Major
Mechanical Energy Engineering
GPA
4.1 / 4.5
Coursework
Linear Algebra, Computer Programming, Data Structures, Databases, Data Processing & Analysis
Awards
Dean's List (Spring 26`- 1st of Class)

Work Experience

IBM

AI Engineer Intern, Client Engineering Team

Seoul, Korea

Jul. 2026 - Sep. 2026
  • Multi-AI Agent System for Enterprise Applications

    Built RAG-powered multi-agent workflows and automated trace-based LLM-as-a-judge evaluations to standardize agent performance assessment.

  • IT Asset Management Dashboard Development

    Built an IBM Instana-based dashboard for 2+ client PoCs and product implementations; batch API requests cut initial loading time by over 90% and asset collection time by over 40%.

  • Technical Seminars / Conferences

    Presented IBM software solutions to 10+ client representatives at technical seminars and conferences, arranging follow-up product adoption meetings.

CAU Energy Conversion System LAB (ECSL)

Research Intern

Seoul, Korea

Jul. 2025 - Oct. 2025

Multi-Stage Restaurant Recommendation System

Dec. 2025 - Present

Developed a two-stage restaurant recommendation system using self-crawled data of over 90K+ user experiences, studying ways how to improve personalized restaurant recommendation quality.

LLM Personalization & Evaluation via Activation Steering

Sep. 2026 - Present

Exploring training-free personalization of Qwen3-0.6B using ratings and survey data from the PRISM user dataset, applying activation steering across hidden layers and evaluating responses through LLM-as-a-Judge.

Python PyTorch HuggingFace Qwen3 0.6B OpenRouter Google Colab

News Simplification for Youth via LLM Fine-Tuning

Mar. 2025 - Jul. 2025

Fine-tuned Gemma 3-1B with QLoRA on 15,894 Korean text pairs synthesized from NIKL using GPT-4o. Simplified news for younger readers while fully preserving the original numbers, amounts, and ratios in 75%+ of samples.

Python PyTorch HuggingFace Gemma 3 1B TRL/PEFT NLP_Newspaper_CUAI

URL-based AI Search Visibility (GEO) Diagnosis Platform for Hotels

Dec. 2025 - Mar. 2026

Developed a URL-based GEO audit platform with Kaicore Corp, enabling 10+ prospective clients to assess hotel websites and receive automated AI-search visibility reports. Built review crawlers and modernized the agency website's UI/UX and SEO/GEO settings using MCP workflows.

Front: React TypeScript Vite Tailwind CSS Back: Node.js Playwright SQLite Resend Docker

Global Supermarket Retail Loss & Business Dashboard

Mar. 2025

Analyzed customer, product, market, order, and shipping data from 50K+ global retail transactions, identified key drivers of financial losses, and proposed three business improvement strategies, including discount policy optimization and product portfolio management.

Hackathon

Codex Community Hackathon — Campus Mate

AI-powered campus lunch pairing service

Aug. 2026

Built an Lunch matching service to connect students based on their interests using Express/TypeScript. Selected as one of the Top 5 services at a hackathon hosted by OpenAI.

Fraudulent Account Detection with Ensemble Learning

Internal data competition — 1st Place (Grand Prize)

May 2025

Led the team to 1st Place (Grand Prize) with a stacked XGBoost/CatBoost ensemble and Optuna tuning, improving Average Precision (AP) by over 20% against the baseline.

Python XGBoost CatBoost Optuna