JIWOONG KIM

Seoul, Korea | jkimak1124@gmail.com | linkedin.com/in/jiwoong-kim-b9934417a

Summary

Quantitative Researcher (Ph.D. Physics) who conceived and drove a self-directed three-phase research agenda: run one ML model live with real capital → industrialize multi-style portfolio generation → automate the research process itself with a multi-agent system. Every phase closed with externally verifiable evidence — a published fund track record, an institution's due-diligence-backed MOU, and controlled experiments on the research harness itself (p = 0.004). Seeking to bring scalable, AI-native alpha research capability to a systematic investment platform.

Experience

Quantitative Researcher | Alpha Bridge (formerly Asset Plus AM), Seoul | Mar 2024 – Present

Conceived, proposed, and led a three-phase initiative to evolve the firm's research infrastructure from one live model to an autonomous research system.

Phase 1 — Launched Firm's First ML-Based Public Mutual Fund

• Built and operate a tree-based fundamental alpha model for a publicly offered US equity mutual fund — concentrated ~30-name S&P 500 universe, long-only, unlevered.
• Live track record since Sep 2024 (published fund fact sheet, as of Aug 2026): +53.1% cumulative vs +35.4% benchmark (+17.7pp); 1-year +29.4% vs +22.2%, top-quartile peer rank.
• Jensen's alpha +6.9%/yr at 0.77 beta (trailing 1-year); beat the benchmark in 8 of 12 months, including 4 of 5 benchmark-down months.
• Run the full production loop end-to-end: signal generation → portfolio construction → execution support → performance monitoring.

Phase 2 — Multi-Style Portfolio Generation Pipeline & Institutional Validation

• Built an automated pipeline that generates diverse financial-factor-based target labels, trains prediction models for each, and backtests to populate the risk-return plane — enabling clients to select portfolio styles that match their preferences.
• Secured a strategy-supply MOU with Daol Asset Management following external due diligence.

Phase 3 — Automating Research Itself: Agentic Alpha Factory (Team Lead)

• Leading a 2-person team building a multi-agent alpha factory that automates idea generation → code generation → backtesting → signal evolution, producing diversified, low-correlation alpha candidates at scale.
• Built a recursive self-improvement loop: every outcome, including failed experiments, feeds back into the next research cycle. Verified by controlled ablation — same base model (Claude Fable 5) with vs. without the loop, graded blind on realized outcomes: 3.6/10 vs 6.9/10 (chance ≈ 2.5), permutation test p = 0.004.
• Automated generation of 1,000+ strategy ideas per day with alpha synthesis into composite signals.
• Composite alpha from the factory — 6 interpretable fundamental signals, long-only, unlevered: IR vs S&P 500 held out-of-sample, 1.06 in-sample (2010–23) → 1.11 OOS (2024–26) → 1.18 including a 4-month live paper-trading period; OOS CAGR 28.6% vs 15.9% benchmark.

Education

• Ph.D. in Particle Physics | Kyungpook National University | 2019 – 2024
• M.S. in Particle Physics | Kyungpook National University | 2017 – 2019
• B.S. in Physics | Kyungpook National University | 2013 – 2017

Research: Signal Extraction Under Extreme Noise (CERN LHC)

• Pioneered CNN-based classification at a collaboration that relied on traditional physicist-driven methods, achieving 1.85x signal efficiency improvement.
• Published results at ACAT 2021 and awarded Best Paper at Korean Physical Society (2020) and Korean Society for Computational Science & Engineering (2021).
• Distributed deep learning across 1,024 compute nodes (KISTI Nurion HPC).
• Publication: J. Phys.: Conf. Ser. 2438 012103 (ACAT 2021 proceedings)

Skills

• Languages: Python, SQL, Shell (Linux/macOS)
• ML / Stats: XGBoost, PyTorch, scikit-learn, pandas, polars, NumPy
• AI Agent: Harness engineering (Claude Code, OpenClaw)
• Infrastructure: Git, Docker