About
“I believe in the brightening and cozy future, led by cognitive computing.”
My journey into ML started with curiosity about the brain — how we think, learn, and perceive the world. At KAIST, I studied Bio & Brain Engineering with a minor in Computer Science, and had the great fortune of being mentored by Prof. Sang Wan Lee at the Lab for Brain & Machine Intelligence. Under his guidance, I explored the intersection of neuroscience and AI — from deep RL for EEG signal processing to multi-agent cognitive policy learning — which shaped the way I think about intelligent systems today.
That academic foundation led me to co-found Humelo, a speech/audio AI startup where I wore many hats (Research Lead, COO, Co-Founder), published at ICASSP, showcased AI music at SXSW, and learned how to build a company from scratch. From there I moved to RIIID, where I built MLOps infrastructure and achieved SOTA on student modeling, before joining Bucketplace (오늘의집) — Korea’s largest interior & lifestyle platform — where I now work across Search (ranking, multimodal retrieval, A/B experimentation) and AIL (Applied Intelligence Lab — agentic AI, GenAI workflows, and 3D scene understanding).
Over 7+ years and three companies, I’ve shipped production ML systems end-to-end and hold 8 registered patents with 3 more applied. I also enjoy bridging industry and academia — 6 papers at international conferences (CIKM, InterSpeech, ICASSP, IEEE SMC) with 5 more at domestic venues and arXiv.
Beyond R&D, I enjoy inventing, prototyping, mentoring, and creating music.
Core Tech
Programming: Python (primary), SQL · Kotlin, Go, C++, Java, JavaScript, Scala, Dart
ML/DL & AutoML: PyTorch, TensorFlow, HuggingFace, Lightning AI, XGBoost, LightGBM, LoRA, Optuna, Ray Tune
Agentic AI & Protocols: LangChain, LangGraph, ADK, LangFuse, A2A, MCP, Claude Code, Codex, Cursor
MLOps & Serving: BentoML, Triton, TorchServe, MLFlow, Airflow (K8s), W&B
Data & Search: ElasticSearch, Spark, Athena, BigQuery, Redis, PostgreSQL, MongoDB, Milvus
Web/App: FastAPI, Django REST, React, Flutter
Cloud & CI/CD: AWS (EC2, S3, ECR, RDS, Lambda), GCP (Cloud Computing, GCS), Docker, GitHub Actions
Education
| Ph.D. Student (Withdrawal) | KAIST, Lab for Brain & Machine Intelligence (2019–2022) |
| M.S. in Bio & Brain Engineering | KAIST (2017–2019). Thesis: Attentional Control Methods for Time-series Data Classification and Synthesis |
| B.S. in Bio & Brain Engineering | KAIST, Minor in Computer Science (2012–2017) |
Selected Publications
| Year | Publication | Venue |
|---|---|---|
| 2025 | MOHPER: Multi-objective HPO for E-commerce Retrieval | CIKM (Applied Research Track) |
| 2023 | Addressing Cold Start for E2E Speech Scoring | InterSpeech |
| 2020 | Multi-speaker Emotional Voice Conversion (FHVAE) | ICASSP |
| 2019 | Phonemic-level Duration Control for Speech Synthesis (oral) | ICASSP |
| 2019 | Polyphonic Sound Event Detection with Transfer Learning | ICASSP |
| 2018 | EEG Signal Classification via Deep RL (oral) | IEEE SMC |
Plus 5 more publications, 2 arXiv preprints, and a Best Paper Award (KIIS 2016).
Patents (8 Registered, 3 Applied)
Registered:
| # | Title | Registration |
|---|---|---|
| I | Adaptive EEG signal processing using reinforcement learning | KR 1023187750000 |
| II | Voice conversion system and method | KR 1022772050000 |
| III | System/device/method to generate polyphonic music | KR 1022274150000 + PCT |
| IV | Training sound event detection model | KR 1021724750000 |
| V | Training sound event detection model | KR 1020256520000 + PCT |
| VI | Apparatus for synthesizing speech | KR 1020579270000 + PCT |
| VII | Apparatus for synthesizing speech | KR 1020579260000 + PCT |
| VIII | Vibration sensor | KR 1012602650000 |
Applied:
| # | Title | Application |
|---|---|---|
| I | E-commerce retrieval support | KR / US / JP |
| II | Supervised contrastive learning for patch-level temporal representation | KR |
| III | Training sound event detection model | KR |
Work Experience
Bucketplace (오늘의집) — ML Engineer Jan 2023 – Present
Search & Applied Intelligence Lab (AIL)
On the AIL side, I’ve led ML-side development of GenAI interior design workflows that achieved +253% Net Satisfaction, contributed to scene-to-products retrieval for Digital Twin with 13x recall improvement (enabling 6000x cost reduction in Image-to-3D), and I’m currently building an agentic natural language compositional retrieval system for RoomPlanner using LangGraph and A2A. I also built a multi-agent R&D workflow automation plugin (17 agents, 14 skills, cross-model code review).
On the Search side, I’ve architected and shipped 11 A/B experiments (8 winners) spanning multi-objective ranking HPO (SERP CTR +0.95%, published at CIKM'25, patent applied), deals ranking (buyer conv up to +11%), and CLIP-based hybrid retrieval (Query CTR +3.03%, CTCVR +16%). I also built the team’s ML experiment infrastructure — offline evaluation, log monitoring, and automated A/B analysis.
RIIID (뤼이드) — ML Engineer, Research Scientist Jul 2020 – Jan 2023
MLOps & AI Research
I shipped and operated model registries, dataset pipelines, and distributed training infrastructure serving 4+ EdTech products (SANTA TOEIC, IVYGlobal SAT, CASA GRANDE, INICIE), improving GPU utilization from 25% to 95% with the company’s first multi-GPU training setup.
On the research side, I achieved SOTA on student modeling (dropout prediction, knowledge tracing) via Attentive Conditional Contrastive Learning (ACCL) across 6 benchmarks, and developed a multimodal speech scoring system deployed to SANTA Say (TOEIC Speaking), published at InterSpeech 2023. I also presented at PyCon KR 2021 on multi-domain ML deployment.
Humelo (휴멜로) — Research Lead; COO; Co-Founder Apr 2018 – Jun 2020
Speech/Audio AI Startup
I co-founded Humelo and managed 5+ AI R&D projects — Emotional TTS (ICASSP 2019, oral, 1st author), Voice Conversion (ICASSP 2020), Sound Event Detection (ICASSP 2019), Speech Emotion Recognition, and AI Music Composition (showcased at SXSW 2019, featured on KBS Documentary). Published 3 top-venue papers as 1st or corresponding author.
As COO, I handled HR, finance, and culture, and secured ~$875K across 3 government R&D grants (IITP, TIPS, Seoul R&BD). The company grew from founding through its Series A stage.