Sean Chang / Research Engineer, Google Research · Health AI

Bringing better health answers to everyone.

I build the models, safety systems and infrastructure behind health answers at Google. On the side, I'm building Corneo, a skincare routine app.

Based in New YorkLast update · Oct 9Agent-readable · /llms.txtLinkedIn ↗
01 What I do

From model research to the systems that serve it.

Health AI safety

Classifiers and evaluations that decide when a medical question or answer needs extra care.
Safety classifiersEvaluationClinical grounding

LLM systems and agents

Retrieval that grounds answers in search and literature, and memory that lets agents hand off context.
RAGAgent memoryGrounding

Model training and research

Fine-tuning and controlled experiments, from hypothesis to a measured result.
Fine-tuningLoRA / QLoRAExperiment design

Cloud infrastructure at scale

Production services on GCP and AWS that stay up when real traffic arrives.
GCPAWSDistributed systems

Products, end to end

Taking an idea to a shipped app on my own: mobile, web and the backend behind them.
iOSWebBackend
Languages: PythonC++JavaTypeScriptSQL

I like building the whole path: the experiment that proves a model is good enough, and the systems that serve it safely at scale.

02 Work at Google

Health AI and Search, at scale.

dosagequestion?query checkGeminiresponse checksafeextra care
Product2025

AI Mode in Google Search

Query + response safety classifiers for medication dosage

Search's conversational mode, built on Gemini. I built the query and response classifiers that keep answers about medication dosage safe.
before visitin visitafter visitshared memorywriteread
Product2025

Fitbit Plan for Care

Infrastructure + shared inter-agent conversational memory

A Fitbit Labs experiment that helps people prepare for doctor visits. I built its infrastructure and the conversational memory its agents share.
Google Searchmedical literatureGeminiRAG harnessconsumer answer12
Research2024

Med-Gemini

Consumer answers grounded in Search + medical literature (RAG)

Gemini models adapted for medicine. I studied answers written for consumers, grounded in fast Google Search and medical literature through a retrieval (RAG) harness.
03 Explainer

Why fine-tuning with LoRA is so cheap.

Two thin matrices instead of one huge one. Drag the sliders.

Trainable share of this layer0.78%
Full update, d²16,777,216
LoRA, 2·d·r131,072
ΔW (d × d)≈B (d × r)A (r × d)

Rank widths are drawn 4× wider than scale so small ranks stay visible.

04 Ask this site

Retrieval you can watch.

BM25 over this page, in your browser. No LLM, so it only quotes me. Tune it in the lab ↗

  1. 1Tokenize
  2. 2Score (BM25)
  3. 3Rank top 3
  4. 4Quote best passage
Query tokens (stemmed, stop words removed)
aimode
Answer, quoted from the best passage

AI Mode in Google Search (2025, product at Google). Search's conversational mode, built on Gemini. I built the query and response classifiers that keep answers about medication dosage safe.

  • 5.96

    AI Mode in Google Search (2025, product at Google). Search's conversational mode, built on Gemini. I built the query and response classifiers that keep answers about medication dosage safe.

  • 1.32

    Health AI safety. Classifiers and evaluations that decide when a medical question or answer needs extra care. Keywords: Safety classifiers, Evaluation, Clinical grounding.

  • 1.09

    Agents and AI assistants can read this site through https://seanchang.me/llms.txt and a read-only MCP server at https://seanchang.me/mcp with tools for profile, skills, Google work, Corneo, updates and site search.

✦ Pattern

How agents hand off context.

Write facts with provenance, read only what the task needs. A general pattern, not a specific product.

05 Side project Example numbers

Corneo, built in public.

Skincare routines built from what's already on your shelf.

Morningfrom your shelf · 4 stepsAMPMCleanseTreatMoisturizeSPF 50Start routine
iOS app · skincare routines

Corneo

Builds a routine from the products you already own, then adjusts it as your skin changes. Recommendations lean on evidence, not hype. Shipped: Routine builder from owned products.
Waitlist
TestFlight testers
Beta rating
Waitlist, last 12 weeks+82 this week
06 Now

A changelog that writes itself.

Drafted weekly by an agent. Published only when I approve.

my reposallowlistagent draftsweeklyI approvenothing auto-postssite + feed/api/now · RSS

Rebuilt seanchang.me with GSAP, Lenis and React Bits.

git · seanchang.me

Corneo: finished the routine builder that starts from products you already own.

git · corneo-ios

Also available as JSON at /api/now.

07 Agent-native

Readable by your assistant, too.

Send your AI first. The tabs hit the same live endpoints it would.

  • /llms.txt summary for any model
  • A read-only MCP server at /mcp
  • Exposes only what is already on this page
Point your own assistant at it
claude mcp add --transport http seanchang https://seanchang.me/mcp
$ curl https://seanchang.me/llms.txt
# Sean Chang
> Research Engineer, Google Research · Health AI. Bringing better health answers to everyone.

I build the models, safety systems and infrastructure behind health answers at Google. On the side, I'm building Corneo, a skincare routine app.

## Core skills
- Health AI safety: Classifiers and evaluations that decide when a medical question or answer needs extra care.
- LLM systems and agents: Retrieval that grounds answers in search and literature, and memory that lets agents hand off context.
- Model training and research: Fine-tuning and controlled experiments, from hypothesis to a measured result.
- Cloud infrastructure at scale: Production services on GCP and AWS that stay up when real traffic arrives.
- Products, end to end: Taking an idea to a shipped app on my own: mobile, web and the backend behind them.
- Languages: Python, C++, Java, TypeScript, SQL

## Work at Google (public)
- [AI Mode in Google Search](https://blog.google/products/search/ai-mode-search/) (2025): I built the query and response classifiers that keep answers about medication dosage safe.
- [Fitbit Plan for Care](https://support.google.com/fitbit/answer/14566053?hl=en) (2025): I built its infrastructure and the conversational memory its agents share.
- [Med-Gemini](https://research.google/blog/advancing-medical-ai-with-med-gemini/) (2024): I studied answers written for consumers, grounded in fast Google Search and medical literature through a retrieval (RAG) harness.

## Side project
- [Corneo](https://corneo.app): Builds a routine from the products you already own, then adjusts it as your skin changes. Recommendations lean on evidence, not hype.

## Links
- LinkedIn: https://www.linkedin.com/in/seantc/

## Machine access
- MCP (Streamable HTTP): https://seanchang.me/mcp
- Latest updates (JSON): https://seanchang.me/api/now
- Updates feed (RSS): https://seanchang.me/feed.xml