Applied NLP / AI Researcher · Seoul, Korea

Grounded language systems, backed by evidence.

I work across retrieval, evaluation, model adaptation, and serving. Earlier research on how brains and language models represent meaning informs how I approach grounding, faithful measurement, and abstention.

See work and run it GitHub Email

Applied NLP
grounded QA · retrieval · embeddings
Research to systems
evaluation · adaptation · serving
Public evidence
code · models · papers · limitations

Selected evidence

Start here

Each item states what is public, what can be run, and what remains private.

Case study

Grounded QA for enterprise contact centers

A verified-unit QA design that returns an operator-approved answer or abstains instead of generating unsupported policy text.

The documented held-out evaluation reports hybrid retrieval with query augmentation at approximately R@1 0.881–0.930.

Evidence →

Public model + evaluation

Korean retrieval embeddings

Two public Korean retrieval encoders with model cards and documented AutoRAG evaluation.

On the documented 720-item, 114-query AutoRAG evaluation, the BGE-M3 variant reports MRR 0.7773 and Hit@10 0.9474.

Evidence →

Runnable code

VisionCardio

An on-device rPPG research prototype spanning model training, Core ML export, and a SwiftUI application.

With a strict participant split, the documented UBFC validation heart-rate MAE improved from 5.63 to 2.80 bpm after fine-tuning.

Evidence →

Evidence policy. Client names, proprietary code, raw logs, and private data are not portfolio material. A case study is labeled as a case study; runnable code is given a concrete smoke command; model claims link to their model cards and evaluation scope.