cv
General Information
| Full Name | Jehyeok Yeon |
| Date of Birth | 16th October 2003 |
| Languages | English, Korean |
| Contact | jehyeok2@illinois.edu |
| https://www.linkedin.com/in/jehyeoky | |
| Programming Languages | Python, C++, SQL, Java, HTML/CSS/JavaScript, C#, Temporal |
| Frameworks and Tools | React, LangChain, ElasticSearch, PyTorch, OpenCV, TensorFlow, Flask, Django, Git, Docker, Streamlit, MySQL, Google Cloud Platform, RESTful APIs, ChromaDB, Mlflow, Spring Boot |
Education
-
August 2022-December 2025 B.S. in Computer Science + Linguistics
University of Illinois Urbana-Champaign - GPA - 3.92
- Dean's List
- Clubs
- Codable (President), ACM, Project Code
- Coursework
- Machine Learning, Data Mining, Database Systems, Robot Manipulation, Corpus Linguistics, Computational Linguistics, Algorithms, Data Structures, Compilers, Discrete Structures, Computer Systems, Senior Project, Statistics and Probability I
Research Interests
- Trustworthy AI
- Explainable AI
- Formal Methods
- Generative AI
- Multimodal Models
- Agentic AI
Publications
- {"Hangoo Kang*, Jehyeok Yeon*, Gagandeep Singh. **TRAP"=>"Targeted Redirecting of Agentic Preferences.** NeurIPS 2025. (*indicates equal contribution)"}
- Jehyeok Yeon, Isha Chaudhary, Gagandeep Singh. **Certifying Robustness of Agent Tool-Selection Under Adversarial Attacks.** (Under Review)
- {"Jehyeok Yeon, Yifan Wu, Federico Cinus, Luca Luceri. **GSAE"=>"Graph-Regularized Sparse Autoencoders for Robust LLM Safety Steering.** (Under Review)"}
- {"Jehyeok Yeon, Lawrence Angrave. **The Power of Friendship"=>"Analyzing Leadership and Adversarial Attacks in Multi-Agent Collaboration.** ACM Collective Intelligence 2025 Poster Acceptance."}
Relevant Experience
-
January 2026 - August 2026 Visiting Researcher
Max Planck Institute for Intelligent Systems - Will be conducting research at the AI Safety and Alignment group under Maksym Andriushchenko about understanding and improving safety of computer use AI agents.
-
November 2024 - Current Research Assistant
FOCAL Lab@UIUC - Achieved **100% ASR** on SoTA vision-language models via novel embedding-level semantic injection and diffusion decoding (first-author, advised by Prof. Gagandeep Singh).
- Developed the first statistical certification framework for agentic AI tool selection under adversarial scenarios via LLM-based adversarial distributions (first-author, advised by Prof. Gagandeep Singh).
- Conducting in-depth research on the impact of demographic and social biases in agentic AI systems deployed on web-based graphical user interfaces (GUIs).
- Designing and executing experiments to uncover and optimize potential avenues for adversarial exploitation.
-
February 2025 - Current Research Assistant
ISI@USC - Designed a graph-based analysis method to uncover features related to LLM refusal behavior using graph Laplacian regularization on sparse autoencoders (first-author, advised by Prof. Luca Luceri).
- Built a steering mechanism that uses a dual gating system with hysteresis to enable control over safety behaviors while maintaining strong utility, performing **40% better** than previous safety steering methods.
-
May 2025 - August 2025 Machine Learning Intern
Intradiem - Trained and tuned a feedforward neural network to predict agent burnout and attrition; optimized for imbalanced classes, achieving **17% F1 lift** and robust generalization to unseen shifts.
- Deployed an agent burnout and attrition prediction model as Spring Boot API; served **12k QPS** at <180 ms latency, automating retraining and drift detection using Temporal and Mlflow for 99.95% uptime.
-
June 2024-August 2024 AI Data Scientist Intern
Hanwha Life Insurance - Developed a hybrid semantic-lexical retrieval architecture for complex tabular data using ElasticSearch and ChromaDB, improving recall on legal document QA datasets by **427%**.
- Built a retrieval augmented generation chatbot by integrating a cross-encoder reranking, multi-step query decomposition, and Hypothetical Document Embeddings, raising MRR by **38%**.
- Optimized query performance and accuracy by implementing advanced RAG strategies including reranking, query defragmentation, and Hypothetical Document Embeddings (HyDE).
-
May 2023-August 2024 Machine Learning Intern
ATLAS - Developed a high-fidelity forecasting system using real-time Chicago city sensor data, fusing multi-modal IoT streams and high-res imagery via LSTM-TCN hybrids in PyTorch; achieved a **27% reduction in RMSE** and 0.91 F1 on severe event prediction.
- Engineered cross-modal attention for feature alignment and validated robustness with adversarial stress tests, maintaining **93% accuracy** under noisy sensor dropout and outperforming baseline uni-modal models by **18%**.
- Processed over 100,000 rows of Canadian agriculture data to analyze temporal conditions' impact on wheat yield, contributing to improved crop management strategies.
-
2024-current Research Assistant
UDL and Accessibility Research Group - Evaluating effectiveness leadership and organizational theories in enhancing collaboration and performance of multi-agent AI frameworks under adversarial conditions.
- Analyzing AI agents' ability to detect and mitigate subtle malicious injections by integrating human-inspired leadership dynamics.
-
2023-2024 Computational Research Assistant
Political Ideology and Entrepreneurship Lab - Engineered automated Python tools for collecting, quantifying, and analyzing data related to political ideology of CEOs and their respective earnings reports.
- Conducted longitudinal research to quantify the political ideology of CEOs using publicly available data.
- Developed and applied novel algorithms to classify CEOs along the changes in political spectrum over time.
-
2023-2024 Data Analyst
University of Illinois Urbana-Champaign, Illinois Leadership Center - Conducts data analysis using Excel and Tableau to extract actionable insights.
- Produces data reports, visualizations, and presentations for ILC workshops and programs.
Ongoing Projects (To be submitted in 2026)
- {"Jehyeok Yeon, Isha Chaudhary, Gagandeep Singh. **When Context Breaks Representation"=>"Understanding Layer-Input Failures in LLMs** (Working Title)."}
- {"Jehyeok Yeon, Federico Cinus, Luca Luceri. **Temporal Tomography of Conceptual Learning"=>"An SAE-based Analysis of LLM Pre-training** (Working Title)."}
- {"Jehyeok Yeon, Hyeonjeong Ha. **All Roads Flow to Rome"=>"Securing Multimodal Models Against Cross-Modality Attacks** (Working Title)."}
Relevant Projects
-
Current Deetox
- Developing a mobile app promoting intellectual growth and digital detox through curated lessons in philosophy, art, music, and literature.
- Delivering a seamless cross-platform experience using Flutter (iOS/Android) and Firebase for real-time data synchronization and user authentication.
- Automating daily lesson generation with large language models (LLMs), optimizing content generation workflows.
-
2024-2024 Workout Planner Application (Eclipse)
- Developed a full-stack workout planning application using React, Node.js, and MySQL, deployed on Google Cloud Platform with RESTful APIs for efficient data communication.
- Implemented comprehensive CRUD operations for workout routines, exercises, diets, and user profiles, ensuring seamless data management and real-time synchronization.
-
2024-2024 ClassTranscribe
- Engineered React components and C# .NET Core backend services to enhance accessibility for visually impaired students.
- Implemented full-stack solutions to transform lecture videos into accessible, picture book-style content, enhancing user experience for visually impaired students.
-
2023-2024 ESL Learning Hub Web App (Grammaraide)
- Designed and implemented a Flask-based web application with responsive frontend using HTML5, CSS3, and JavaScript to enhance ESL students' reading comprehension and grammar proficiency.
- Integrated advanced NLP methods and LLM pipeline to generate personalized, contextually relevant quizzes and exercises, catering to various ESL proficiency levels.
-
2023-2023 Note Generation Chrome Extension
- Developed a Chrome extension utilizing Flask backend and JavaScript frontend, integrating the open-source LLM Mixtral and OpenAI's wav2vec 2.0 speech-to-text model.
- Achieved 85% accuracy in automated note generation from YouTube videos and audio files, enhancing user productivity by over 10 hours per week.
-
2022-2023 Chopstick 101
- Used TensorFlow and OpenCV to develop a machine learning program that analyzes and rates real-time chopstick holding positions applying LSTM on the hand landmark positions.
Honors and Awards
- Get Experience Scholarship (2023)