Selected work / public mission + private execution
Systems built to survive reality.
Federal research platforms, responsible AI, and production cloud products. The common thread is making clear architecture and delivery decisions under real constraints.
Modernizing the information layer behind regulatory science.
My work at FDA’s Center for Tobacco Products connects research operations, data integration, scientific retrieval, acquisition, and product delivery. Much of the implementation is internal; the mission and several core systems are documented publicly.
CTP Integrated Research Data System
CIRDS
I helped lead CIRDS from proof of concept toward production and am the first author on FDA’s public technical poster about the platform. CIRDS brought previously disconnected tobacco-research sources into a unified search and discovery experience for CTP scientists.
View the FDA technical poster ↗Integrated records
Research projects and documents brought together across four heterogeneous sources.
Source systems
RTS, TCKB, iDAT, and the UCSF Truth Tobacco Industry Documents archive.
Mission context
FDA’s current IT plan describes AIRRS as a CTP Office of Science suite supporting scientific and business analysis.
FDA IT plan ↗02 / AI inside research tracking
Responsible AI, embedded in the operating workflow.
I led the integration of an in-boundary LLM into CTP research tracking applications. It analyzes regulatory research abstracts against the coding manual and returns recommendations, evidence snippets, and discrepancy flags.
The architecture is intentionally human-in-the-loop: AI narrows the review surface; regulatory science staff retain authority.
Reduce low-value comparison work
Direct expert attention toward disagreements and edge cases.
Ground every recommendation
Use the governing manual and surface the evidence behind the output.
Keep reviewers accountable
The system recommends. FDA staff decide.
Design for the agency boundary
Integrate into the existing research workflow and security environment.
03 / Native product / active build
SAA—
Swipe
AWS certification content is abundant; decision fluency is not. SAA-Swipe is designed around the signals that distinguish a plausible answer from the strongest architecture choice.
A study system built around architecture judgment.
The MVP deliberately connects three modes—learn, study, practice—so content is reinforced in context instead of becoming another disconnected question bank.
Problem framing
Shift exam prep from memorizing services toward recognizing constraints, tradeoffs, and decision fingerprints.
Product site ↗Experience model
Keep the daily loop focused: follow the course, practice the clues, and know what needs another pass.
Delivery posture
Build a coherent native product and content pipeline first; add complexity only when product evidence demands it.
OAC
JWT
Lambda
EventBridge
04 / AWS platform / production
PetShots
Pet records with shareable, expiring QR passports. Built as a serverless system to keep idle cost and operational burden close to zero without giving up clean security boundaries.