NLP & AI research
I work with Prof. Heuiseok Lim at the NLP & AI Lab. My research spans private inference, multilingual retrieval, and agent evaluation.
I am interested in making model reasoning more efficient: using less computation while preserving answer quality, and extending this work to multimodal models.
Open to research internships and collaborations.
* Equal contribution.
A benchmark of 1,011 scientific questions that tests how agents find and combine evidence across papers. It diagnoses retrieval, synthesis, tool use, and resource management.
My role: Contributed ideas, assisted with experiments, and helped write the paper.
PPFT lets a server-side LLM answer from noisy prompt embeddings without receiving the raw text. We study the trade-off between answer quality and resistance to text reconstruction.
My role: Led the project from the initial idea through experiments, analysis, and writing.
Uses a sparse autoencoder to remove language-specific features from text embeddings, improving multilingual retrieval without retraining the encoder.
My role: Contributed to experiments, analysis, and writing.
Adds uncertainty estimates to pretrained models through a lightweight transformation of their output logits, evaluated on image classification and LLM question answering.
My role: Contributed to initial experiments and writing.
Tests whether 18 LLMs detect ethically problematic language when it appears in code comments, identifiers, and other parts of Python and C programs.
My role: Contributed to experiments, analysis, and writing.
Aligns text and image representations for medical question answering without sending raw inputs to the server. The server adapts to the domain using noisy representations.
My role: Developed the idea, carried out the experiments, and wrote the paper.
A Korean language-consultation benchmark for expression acceptability and rule grounding. Supplying the correct rules improves citations but does little to improve the underlying judgments.
My role: Contributed to initial experiments and analysis.
Studies how many upper layers can be removed from an LLM used as a text embedder while retaining retrieval quality.
My role: Designed and carried out the study from start to finish.
A Korean-English RAG model trained to decide when to retrieve, filter irrelevant evidence, and check its answers.
My role: Contributed to experiments and analysis.
No publications in this topic.
The server generates references to private input content, and the client resolves them to the original text. This lets a model reuse private strings exactly while reducing plaintext exposure to the service provider.
Uses early voting and prunes unproductive reasoning trajectories to reduce the cost of multi-sample inference without additional training.
My role: Assisted with initial experiments and helped write the paper.
No preprints in this topic.
Graduate researcher at the NLP & AI Lab, advised by Prof. Heuiseok Lim.
Double major in Software Engineering, with project work spanning simulation, autonomous driving, reinforcement learning, and machine learning.
Implemented and debugged finite-element simulation components for beam and shell structures using C++ and MATLAB, including 6-DOF representations and coordinate transformations.
Built perception, planning, and control components using LiDAR, cameras, ROS, MATLAB/Simulink, and Gazebo.
Text embeddings, sparse autoencoders, dense retrieval, RAG, agent evaluation, uncertainty estimation, and privacy evaluation.
Python, PyTorch, Hugging Face Transformers, TensorFlow, LoRA fine-tuning, model evaluation pipelines, data processing, and experiment automation.
C/C++, MATLAB/Simulink, ROS, Gazebo, finite-element simulation, image-processing pipelines, autonomous-driving control, and reinforcement learning.
For research internships or collaboration on language models, efficient reasoning, and multimodal learning:
aa007878@korea.ac.kr
