Mathematics + Computer Science @ KAIST · IMO Bronze '20 '21 ·ML @HayanMind · AI @Huawei · Computational Mathematics @IBS
I work at the intersection of mathematical modeling, systems, and AI — with a bias toward problems where theory actually constrains what you can build, and where you find out quickly whether you were right.
- KAIST — 4th year, Department of Mathematical Sciences · School of Computing. B.S. Mathematics and Computer Science (double major), graduating Dec 2027.
- WorldQuant BRAIN — Quant Research Consultant. Cross-sectional equity alpha for a long-short market-neutral book: fundamentals, analyst estimates, options analytics, volatility surfaces, news/sentiment, alternative data. Gold Medalist, WorldQuant Challenge.
| HayanMind — ML Engineer Intern, R&D Jun–Aug 2026 |
Dynamic pricing on Contextual Thompson Sampling with Transformer sequential encoders. Hawkes processes for rare-event purchase modeling — +46% PR-AUC on sparse data. Off-policy evaluation via counterfactual simulators and regret curves; proposed a cost-aware retraining policy framework. |
| Huawei — R&D Intern, Hong Kong Jan–Mar 2026 |
LLM inference algorithms for hybrid attention/Mamba architectures. CUDA kernels, KV cache management, throughput/latency/memory optimization in C++ across vLLM, SGLang and TensorRT. Owned benchmarking and profiling of the serving path. |
| IBS — ML Research Intern Center for Mathematical and Computational Sciences · Dec 2024–Dec 2025 |
Deep learning for sleep disorder classification from multimodal physiological time series; inductive bias design under limited data. Stochastic modeling of non-Markovian delays in birth–death processes via likelihood methods and physics-informed neural networks. |
| KAIST Users & Information Lab | Cultural NLP benchmark for low-resource languages; evaluated chain-of-thought breakdown across language contexts. Paper accepted at AACL-IJCNLP 2025 Workshop · ACL Anthology |
| KAIST Data Strategy Lab | Pre/in/post-training bias mitigation in foundation models. |
Volatility Inference with SDEs Heston-lite and OU stochastic volatility models with Kalman and particle filters for crypto vol estimation.
SignalCraft Modular crypto alpha pipeline. ~52% hit rate, Sharpe > 1 with ensemble models and adaptive execution.
Sparse Attention CUDA Kernel Longformer-style sparse attention with Flash Attention online softmax. 1.4–1.8× speedup, O(n·w) complexity, O(1) per-thread memory. Benchmarked against dense PyTorch attention on T4.
Yut AI Minimax agent with Bayesian-optimized policy weights for the Korean board game Yutnori. 56.5% win rate over baseline.
Multi-Agent AI Wellness Platform — funded by Samsung Insurance Lifenology Lab Led a cross-functional entrepreneurial team; owned technical direction. LangGraph multi-agent orchestration with RAG under human-in-the-loop decision constraints.
- Bronze Medalist, International Mathematical Olympiad (IMO) — 2020, 2021
- Bronze Medalist, Asian Pacific Mathematical Olympiad (APMO) — 2021, 2022
- Jane Street Prize · Best-in-University, Simon Marais Mathematics Competition — 2024
- Bronze Medalist, Bangladesh Olympiad in Informatics — 2022
- Gold Medalist, WorldQuant Challenge
- Dean's List (Top 2%) · KAIplus Scholar · ISSS Scholar
- Top 0.1% rapid on Chess.com
Python C++ CUDA C F# Scala SQL JavaScript
PyTorch NumPy pandas SciPy scikit-learn vLLM SGLang TensorRT LangGraph
PostgreSQL Docker FastAPI Django