Xiao Chen
AI, health, and social science research · Former Amazon software engineer
Seeking PhD opportunities · Fall 2027
My work spans distributed systems for pharmacy operations at Amazon Health, large-scale education research, and PhoneMood, an Android research instrument I built alone.
Selected work
Applied AI systems
My role: Built pipeline components and comparison logic for a system that checks prescription images, extracted label text, order records, and database entries against each other before a pharmacist decides.
Amazon Health · pipeline handles 3,000+ prescription images per day at under 800 ms average latency
Full-stack engineering
My role: Architected the review application end to end — interface, APIs, and the workflow state that makes every verification, exception, and reversal auditable.
Amazon Health · 200+ internal users, 10,000+ prescription review requests per day
Independent research instrument
My role: Sole author: sampling protocol, Kotlin and Compose implementation, on-device statistical engine, and an export that carries the records behind every estimate.
Available on Google Play · source code and literature review public
Quantitative research
My role: Credited with methodology, software, formal analysis, and investigation on the PISA 2022 preprint; implemented and compared the predictive models in the STEM career-interest study.
PISA 2022: 19,779 students in 996 schools · STEM study: 15,847 students, 75 predictors
Technical expertise
Design and build a complete application: the interface people work in, the APIs behind it, and the data model that keeps its state consistent.
React · TypeScript · Java · Python · FastAPI · REST APIs
Run services in production — containerized APIs, event-driven updates, caching for read latency, and infrastructure defined in code.
AWS ECS · Lambda · SQS · DynamoDB / DAX · Lightsail · CDK
Put multimodal models into a real workflow: coordinate specialist agents, structure their outputs into a cross-check, and analyze where the system is wrong or unsure.
AWS Bedrock / Claude · Python · agent orchestration · evaluation design · error analysis
Build an Android app that records behavior in the background, issues prompts on a defined sampling rule, and keeps responses and non-responses on the device.
Kotlin · Jetpack Compose · Room · UsageStatsManager · WorkManager
Prepare large survey datasets, fit multilevel and within-person models, compare predictive models, and interpret them for coauthors who do not read code.
R (lme4, lmerTest) · Python (scikit-learn, XGBoost, SHAP) · multilevel modeling · ROC-AUC · effect sizes
Turn raw records into an analysis sample, then export the observations, inclusion decisions, model inputs, and estimates together so a reader can reconstruct the result.
Sample construction · quality metadata · JSON export schema · deletion-based stability checks
Featured research
In the PISA 2022 study where I am credited with methodology, software, and formal analysis, pre-sleep digital self-regulation rests on a single self-report question. The paper’s own limitations section proposes adding behavioral measures such as smartphone logs.
Published comparisons of reported and recorded use treat remembered usage as a measurement to validate, and brief repeated checking is the hardest part to reconstruct afterward. That is a measurement question before it is a substantive one.
PhoneMood records active use, requests a mood rating once a configurable amount of it accumulates, and keeps unanswered prompts, response delays, and monitoring gaps in the analysis sample rather than discarding them.
Also built
A cost-estimation service for fulfillment routing: parallel reads behind a cache, event-driven updates to each facility’s processing path, and explicit unknown-path results so missing operational records stay visible downstream.
A bilingual VPN product I built and still operate: account registration and recovery, subscriptions and payment callbacks, and a scheduled job that reconciles traffic usage from the VPN management system with account records.
A synonym study app with spoken word cards, example sentence audio, Chinese etymology notes, and separate daily new-word and review sessions.
Selected manuscripts
Preprints.org, 2026
My contribution: Methodology, software, formal analysis, and investigation.
Manuscript · 2026
My contribution: Data preparation, implementation and comparison of the predictive models, and SHAP-based interpretation.
Contact
I am looking for a group where measurement design, quantitative analysis, and building the software that collects the data are part of the same job. If that describes your lab, I would be glad to hear from you.