# Jay Kang

New York, NY · jianyu.mfe@gmail.com · https://www.linkedin.com/in/jaykang17

AI specialist with 10+ years across finance and machine learning. I build systems, coordinate stakeholders, and own delivery through production support. I prefer owning direction and leading development while staying close to the technology.

## Current work — Mizuho

External AI specialist consultant via Synechron, December 2025–present.

### Metadata governance platform

I built a production platform used by four domain teams across 10+ products, with hundreds of attributes, thousands of enumerations, and 100+ metadata versions. It connects business meaning to physical tables and columns, supporting reviewed mappings for downstream data transformation and delivery.

The MD supplied the framework; I built out the application as the sole human developer using Copilot and coding agents. I coordinate business users, data teams, the DQ team, and four human QA testers. I handle release work through Jenkins and ongoing production support.

Capabilities include semantic versioning, draft-to-publish approvals, audit snapshots, domain entitlements, maker-checker controls, mutation locks, no-op protection, and configurable DQ metadata. The DQ team owns the rules.

The natural-language interface uses Sonnet and LangGraph and is in beta. Queries can read directly; consequential changes require impact assessment and human confirmation.

**Decision:** Launch the working Streamlit application under a tight schedule and defer an AppKit UI rebuild.

**Evidence limit:** Adoption is established; no measured time-saving percentage is supplied.

### Parser agent

I built a production agent that generates Python parsers and loaders from provider format documentation and feeds reviewed code into daily Databricks ingestion. I own the landing-to-bronze workflow.

Demonstrated scope: ten datasets from one provider in approximately five formats. The agent compares parsed document length with the source and retries mismatches through LangGraph. Humans test and review the generated code before it enters the pipeline.

**Evidence limit:** A length check is not proof of semantic correctness. I do not claim universal format support.

## Earlier work

- **Alysida Capital, September 2021–October 2024:** Cofounder and quantitative trading lead. Hired and trained four researchers and developers, growing to ten including summer interns. Owned hiring decisions as a partner. Worked on reusable model evaluation, leakage and overfitting review, simulations, staged deployment, execution systems, and risk controls. Small firm with proprietary and external capital; no performance claim here.
- **Synovation AI, September 2019–August 2021:** Director of machine learning and strategy systems. Financial ML research frameworks and reinforcement-learning environments.
- **Meritz Asset Management, earlier career, Seoul:** China-focused equity research using SVM screening, company visits, and fundamental reports. A complete employment interval is not supplied in this profile.
- **Jichengli, October 2009–January 2012:** Intraday futures trading.

## IRIS — personal experiment

An experimental memory and retrieval system using spreading activation, motivated by the difficulty of recovering useful context across AI interactions. It is a work in progress; no general performance advantage over RAG is claimed. Project code is separate from the downloadable profile tools.

## Education

- Master of Financial Engineering, Stevens Institute of Technology, 2019.
- Master of Mathematics, Seoul National University, 2014.
- BS Mathematics, University of North Carolina at Charlotte, 2009.

## Tools and languages

Python, SQL, PySpark, Databricks, Unity Catalog, PostgreSQL, Streamlit, LangGraph, Sonnet, Jenkins; additional experience with Databricks SDK/CLI/Jobs APIs, LangChain, JavaScript/HTML, Golang, AWS, and TimescaleDB.

English, Mandarin, Korean.

## What I want next

Ownership of direction and delivery, broader customer problems, and opportunities to turn implementation experience into better products. New York is preferred; open to relocation for the right opportunity and up to 50% travel.

## Topics to clarify in an interview

Commercial ownership of SOWs, budgets, procurement, and contract changes; measured deployment outcomes; evaluation and operational controls; any missing exact employment dates.

---

Updated September 9, 2026. Candidate-provided information edited for clarity; not an independent employment or performance verification.