Hochan Son

(he/him)

Database Architect · SRE & DevOps Engineer · ML/GenAI Practitioner

UCLA Trustworthy AI Lab

Professional Summary

Hochan Son is a distributed systems architect and site reliability engineer who builds data infrastructure, ML pipelines, and high-scale distributed systems, working the gap between GenAI/LLM research and the systems that run it in production. His graduate research at the UCLA Trustworthy AI Lab spans synthetic tabular data generation and agentic, intent-conditioned model selection. His background spans Adtech, entertainment, and enterprise — from MySpace and Hallmark Labs to Branch.io and ADP.

Education

MASDS — Applied Statistics and Data Science

2024 – 2026

University of California, Los Angeles (UCLA)

B.S. CSE — Computer Science and Engineering

2006 – 2009

University at Buffalo (SUNY)

Interests

Generative AI, LLMs, RAG & agentic (Chain-of-Agents) systems Synthetic data generation & privacy-preserving ML Large-scale database engineering (RDBMS, NoSQL, Distributed SQL) SRE, DevOps & cloud-native infrastructure Local LLM inference & distributed GPU training

🔬 Research Focus

I work the gap between GenAI/LLM research and the systems that have to run it in production. Concretely: agentic, intent-conditioned tooling for generative-model selection (Synthony), the training/evaluation infrastructure that supports it at scale (table-synthesizers), and the database, SRE, and DevOps foundation that makes either one reliable.

My thesis direction extends this into applying LLM-driven reasoning — RAG and Chain-of-Agents (COA) orchestration — to systems-engineering decisions, treating “which model,” “which architecture,” and “which infra tradeoff” as the same class of problem: a research-to-production gap that needs both ML judgment and engineering rigor to close.

Publications

Hochan Son

Equal contribution

, Kyungdoe Han

Equal contribution

, Jaehan Koh, Xiaowu Dai, Wenlu Xu, Guang Cheng (2026). Persistent Memory in Multi-Agent LLM Inference: What It Costs, What It Buys, and When You Can Tell. NeurIPS 2026 MLForSys.

PDF Source Document Edge LLM Inference

Hochan Son (2026). Toward an Auto-Synthesizer: Stress-Aware Selection and Recommendation of Deep Tabular Generative Models. UCLA Master’s Thesis.

eScholarship Code Synthony table-synthesizers

Hochan Son, Xiaofeng Lin, Jason Ni, Guang Cheng (2026). SYNTHONY: A Stress-Aware, Intent-Conditioned Agent for Deep Tabular Generative Models Selection. ICLR 2026 DeLTA.

PDF Source Document Code Synthony