Jung Hoon Son, M.D.
Physician-engineer. I build data pipelines, AI agents, ML models, and evidence platforms across healthcare tech — and write about the gap between what models can do and what they actually do.
Selected Work
One Chart, Five Questions
A scroll-driven visual essay: healthcare AI benchmarks are saturated, and the evaluation that answers “does this help?” looks like product analytics — in private.
Question Stream
The precharting funnel rendered as live particles — swarm → the doctor → extraction and reasoning lanes → the usage log.
Working with Data in the Age of GenAI
XPC Hackathon winner. A step-through deck on how analytics work changed — imperative → declarative → English — culminating in a prompt typed into a zsh terminal and a styled chart coming out.
Family History Interface
A structured interface for capturing family medical history.
LOINC Lookup
Type a lab test name, get LOINC codes — Gemini-powered terminology search.
Geomark
A geography quiz game built for my kid — pick countries and watch them fill the map with their flags; hover to learn size, neighbors, and more. Co-authored with Mark Son.
willitrun
Paste a Hugging Face model, find out if the machine you're holding can run it — real parameter counts, KV-cache math, and PASS/TIGHT/FAIL badges in a terminal UI, with a live trending dashboard fit-checked against your RAM.
Marklee
Annotations that travel with the document — a portable, format-agnostic annotation format for PDF, Markdown, and DOCX. Local-first sidecar JSON, edit-tolerant text anchors, MarkRank centrality over your reading.
Experience
- Led product analytics for agentic AI deployments across multiple practice groups — utilization tracking, observability, and clinician-behavior profiling for EHR-embedded products (TypeScript/Python, GitHub/Vercel).
- Guided the main product's transition from rules-based system to AI-native, LLM-centric platform.
- End-to-end development of LLM-based parsing of unstructured clinical notes (PDF, CCDA XML) for agent development in a HIPAA-compliant PHI environment.
- Analytics behind an $8.7B acquisition — built an asset-intelligence platform integrating pipeline data with GlobalData, Evaluate Pharma, and ClinicalTrials.gov; comparative valuation models informed C-suite strategy.
- Built automated clinical-trial monitoring: ETL over Veeva CTMS plus LLM document analysis, with dashboards on GCS / BigQuery / Looker.
- Deployed an AI-powered medical Q&A system (Teams-integrated, GCP + LLMs) for Medical Affairs.
- Predictive models that secured a $16.5M Series A — ML for patient population identification over integrated claims, lab, and clinical datasets; evidence presented to Bain Capital and Two Bear Capital.
- Engineered a chronic-disease analytics platform processing millions of patient records (Apache Spark, Python) for kidney-disease population identification.
- Delivered analytics for enterprise partnerships with US Renal Care and Quest Diagnostics; two patented clinical-data methodologies; presentations at ASN Kidney Week.
- NLM Fellowship, Biomedical Informatics — Columbia University (DBMI), 2016–2017
- Pathology Residency (AP/CP) — New York-Presbyterian / Columbia, 2012–2016
- M.D. — UMDNJ, New Jersey Medical School, 2008–2012
- B.A. Chemistry & Psychology — Cornell University, 2004–2008
Writing
I write at Donuts with Chopsticks on Substack.
-
Model Evaluation as Product Analytics
long-form essay
Capability benchmarks are saturated; the evaluation that answers “does this help?” looks like product analytics, and it happens in private.
Talks & Presentations
-
XPC Talks: How Physicians Hack
in Healthcare
video · X = Primary Care
-
The State of Healthcare Benchmarks
15-slide deck · 2026
-
Chronic-disease analytics and patient identification — podium presentations
ASN Kidney Week · ARVO · AMIA · NLM Annual · USCAP
Patents & Publications
-
Deep Phenotyping on Electronic Health Records Facilitates Genetic Diagnosis by
Clinical Exomes
American Journal of Human Genetics · 2018
-
Interactive bodymap systems and machine learning methods
U.S. Patent 18/219,168
-
Machine Learning Systems and Methods for Predicting Risk of Renal Function Decline
U.S. Patent 16/459,469
-
Monitoring Breathalyzer
U.S. Patent 15/424,771