# Schmiedmayer Lab

> We develop patient-facing AI, multimodal models, and biomedical research agents, and study prevention and digital wellbeing in everyday life.

Canonical site: https://schmiedmayerlab.stanford.edu/
Affiliation: Division of Computational Medicine · Department of Medicine · Stanford School of Medicine, Stanford University
Contact: https://schmiedmayerlab.stanford.edu/contact/

This generated Markdown file is a navigation aid for language models and other agents. The canonical HTML pages and cited primary sources remain authoritative.

## Core pages

- [Home](https://schmiedmayerlab.stanford.edu/): We develop patient-facing AI, multimodal models, and biomedical research agents, and study prevention and digital wellbeing in everyday life.
- [Research](https://schmiedmayerlab.stanford.edu/research/): Enduring research themes and their related outputs.
- [Team](https://schmiedmayerlab.stanford.edu/people/): Current lab members and alumni.
- [Publications](https://schmiedmayerlab.stanford.edu/research/#publications): Filterable bibliography with links to original papers.
- [Projects & resources](https://schmiedmayerlab.stanford.edu/resources/): Models, software, studies, tools, educational resources, and emerging platforms with explicit maturity and availability.
- [Join the lab](https://schmiedmayerlab.stanford.edu/join/): Student, postdoctoral, staff, collaboration, and open-source pathways, current announcements, and contact.
- [Contact](https://schmiedmayerlab.stanford.edu/contact/): Contact information for Schmiedmayer Lab.

## Research

- [Multimodal Computational Methods](https://schmiedmayerlab.stanford.edu/research/#multimodal-health-intelligence): How can language models reason jointly over wearable observations and clinical context?
- [Learning and Evaluation from Patient Interactions](https://schmiedmayerlab.stanford.edu/research/#communication-and-behavior): How can patient interactions guide model training and evaluation?
- [Patient-Centered Agentic Systems](https://schmiedmayerlab.stanford.edu/research/#trustworthy-agentic-systems): How can agents support evolving patient needs under effective supervision?

## Team

- [Paul Schmiedmayer](https://schmiedmayerlab.stanford.edu/people/#paul-schmiedmayer): Instructor, Computational Medicine.
- [Matthias Lutz](https://schmiedmayerlab.stanford.edu/people/#matthias-lutz): Visiting Student Researcher.
- [Karl Aurel Deck](https://schmiedmayerlab.stanford.edu/people/#karl-aurel-deck): Visiting Student Researcher.
- [Michel Schimpf](https://schmiedmayerlab.stanford.edu/people/#leon-michel-schimpf): Former Visiting Student Researcher.
- [Camila Quintero](https://schmiedmayerlab.stanford.edu/people/#camila-quintero): Former Visiting Student Researcher.
- [Linea Schmidt](https://schmiedmayerlab.stanford.edu/people/#linea-schmidt): Visiting Student Researcher.
- [Nicolas Zumarraga](https://schmiedmayerlab.stanford.edu/people/#nicolas-zumarraga): Visiting Student Researcher.
- [Florence Heinkelein](https://schmiedmayerlab.stanford.edu/people/#florence-heinkelein): Visiting Student Researcher.
- [Sophia Trabanino](https://schmiedmayerlab.stanford.edu/people/#sophia-trabanino): Stanford Undergraduate Student.
- [Rieke Hackbarth](https://schmiedmayerlab.stanford.edu/people/#rieke-hackbarth): Affiliated Researcher.
- [Max Rosenblattl](https://schmiedmayerlab.stanford.edu/people/#max-rosenblattl): Affiliated Researcher.
- [Patrick Langer](https://schmiedmayerlab.stanford.edu/people/#patrick-langer): Affiliated Researcher.
- [Thomas Michael Kaar](https://schmiedmayerlab.stanford.edu/people/#thomas-michael-kaar): Affiliated Researcher.
- [Svea Strassburger](https://schmiedmayerlab.stanford.edu/people/#svea-strassburger): Affiliated Researcher.
- [David Grüning](https://schmiedmayerlab.stanford.edu/people/#david-gruening): Affiliated Researcher.
- [Lukas Kollmer](https://schmiedmayerlab.stanford.edu/people/#lukas-kollmer): Software Engineer.
- [Eldi Cano](https://schmiedmayerlab.stanford.edu/people/#eldi-cano): Software Engineer.
- [Arkadiusz Bachorski](https://schmiedmayerlab.stanford.edu/people/#arkadiusz-bachorski): Software Engineer.
- [Paul Johannes Kraft](https://schmiedmayerlab.stanford.edu/people/#paul-johannes-kraft): Software Engineer.
- [Paul Goldschmidt](https://schmiedmayerlab.stanford.edu/people/#paul-goldschmidt): Software Engineer.
- [Milana Gurbanova](https://schmiedmayerlab.stanford.edu/people/#milana-gurbanova): Former Visiting Student Researcher.

## Publications

- [TimeCap: A Scalable Open-Source Pipeline for Captioning Wearable Health Time-Series Data](https://schmiedmayerlab.stanford.edu/research/#rosenblattl2026timecap): TimeCap generates deterministic captions from wearable signals. Applied to OpenMHC, it provides a reproducible caption layer for training and evaluating time-series language models.
- [Language eligibility criteria and enrolled racial and ethnic composition in US clinical trials: a cross-sectional study](https://schmiedmayerlab.stanford.edu/research/#tolas2026language): This cross-sectional study examines how language eligibility requirements relate to the racial and ethnic composition of participants in US clinical trials.
- [PocketTSLM: Deploying an On-Device Soft-Prompt Time-Series Language Model for Privacy-Preserving Smartphone Reasoning](https://schmiedmayerlab.stanford.edu/research/#rosenblattl2026pockettslm): PocketTSLM ports OpenTSLM SoftPrompt to a consumer smartphone and evaluates numerical parity and resource use. Inference remains local; the paper does not establish clinical effectiveness.
- [DrugTargetWorld: A Synthetic Biobank for Training and Benchmarking AI Scientists](https://schmiedmayerlab.stanford.edu/research/#margolis2026drugtargetworld): DrugTargetWorld generates simulated biobanks with hidden causal structure for training and evaluating research agents on drug-target discovery. Its benchmarks use synthetic worlds, not demonstrated clinical discoveries.
- [TimeNet: An Extensible Unified Data Infrastructure for Next-Generation Temporal Foundation Models](https://schmiedmayerlab.stanford.edu/research/#maritsch2026timenet): TimeNet defines a shared representation for temporal signals, metadata, annotations, and task supervision, allowing heterogeneous datasets to be combined for model training and evaluation.
- [OpenTSLM TeeMoE: A Unified Time-Series Language Model for Forecasting, Contextual Prediction, and Reasoning](https://schmiedmayerlab.stanford.edu/research/#chen2026teemoe): OpenTSLM TeeMoE combines specialized low-rank experts for forecasting, contextual prediction, and temporal reasoning in a shared language-model backbone.
- [Beyond Screen Time: Demonstrating the Value of App Activity Logs to Understand User Behavior Further](https://schmiedmayerlab.stanford.edu/research/#schmitt2026beyondscreen): A secondary analysis of event-level app logs examines what measures of everyday smartphone behavior reveal beyond aggregate screen time.
- [How do people plan digitally: An in-the-wild investigation of task planning through a smartphone app](https://schmiedmayerlab.stanford.edu/research/#halder2026planning): An observational study combines smartphone-planner logs and surveys to examine how people create, revise, postpone, complete, and abandon everyday tasks.
- [The CAST-framework: Measure and model social media use as a multi-level phenomenon through real-world applications](https://schmiedmayerlab.stanford.edu/research/#gruening2026cast): CAST connects social-media exposure, behavior, physiology, and self-reported experience across measurement timescales. Its illustrative demonstration uses simulated participants rather than an intervention trial.
- [AuscuTSLM: Patient-Level Multimodal Question Answering from Multi-Site Auscultation Recordings](https://schmiedmayerlab.stanford.edu/research/#wu2026auscultation): Published at Interspeech 2026, AuscuTSLM aligns multi-site auscultation recordings with a language model to investigate patient-level question answering over physiological audio.
- [OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data](https://schmiedmayerlab.stanford.edu/research/#langer2025opentslm): Presented at ICML 2026, OpenTSLM introduces time series as a native language-model modality for reasoning jointly over clinical text and multivariate physiological signals.
- [Spezi Data Pipeline: Streamlining FHIR-based interoperable digital health data workflows](https://schmiedmayerlab.stanford.edu/research/#bikia2026pipeline): A modular Python toolkit supports retrieval, processing, review, and export of FHIR-based digital-health data, with an application to wearable ECG recordings in pediatric research.
- [Bloom: Designing for LLM-Augmented Behavior Change Interactions](https://schmiedmayerlab.stanford.edu/research/#joerke2026bloom): Bloom studies how a language-model coach can augment established behavior-change interactions while incorporating domain-specific safety evaluation.
- [How Well Do Multimodal Models Reason on ECG Signals?](https://schmiedmayerlab.stanford.edu/research/#xu2026ecgreasoning): This work evaluates how multimodal models reason directly over ECG signals and identifies capabilities and limitations relevant to clinical use.
- [Design and Rationale of the My Heart Counts Cardiovascular Health Study: A Large-Scale, Fully Digital Biobank, and Randomized Trial of Large Language Model-Driven Coaching of Physical Activity](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2026myheartcounts): This paper describes a fully digital cardiovascular cohort and an embedded trial of personalized language-model-driven physical-activity coaching.
- [Digital (self-)nudging: A framework for a continuum of user agency](https://schmiedmayerlab.stanford.edu/research/#gruening2026selfnudging): A conceptual framework distinguishing who defines digital intervention options and who chooses among them, with implications for user agency.
- [Grounding Language Models in Behavioral Science to Scale Physical Activity Interventions for Hispanic/Latinx Populations](https://schmiedmayerlab.stanford.edu/research/#mantena2026grounding): A preprint on grounding language-model physical-activity interventions in behavioral science for Hispanic/Latinx populations.
- [OpenMHC: Accelerating the Science of Wearable Foundation Models](https://schmiedmayerlab.stanford.edu/research/#schuetz2026openmhc): OpenMHC provides qualified researchers with access to a large wearable-health dataset, together with reproducible implementations and benchmarks for wearable foundation models.
- [Population-Based Prediction Equations for the 6-Minute Walk Test: Findings From the My Heart Counts Study](https://schmiedmayerlab.stanford.edu/research/#tolas2026sixminutewalk): This preprint develops population-specific prediction equations for unsupervised, smartphone-based six-minute walk tests in the My Heart Counts cohort.
- [The ABC of Digital Health: A Framework for Translating Digital Health Interventions into Real-World Applications](https://schmiedmayerlab.stanford.edu/research/#gruening2026abc): The ABC framework connects accessibility, buildability, and continuity to the practical translation of digital-health interventions beyond trials.
- [TimeRLM: Recursive Language Models Enable Precise Anomaly Localization in Long-Context Time-Series](https://schmiedmayerlab.stanford.edu/research/#zumarraga2026timerlm): An experimental recursive language-model approach to anomaly localization in long time series.
- [TS-Haystack: A Multi-Task Retrieval Benchmark for Long-Context Time-Series Reasoning](https://schmiedmayerlab.stanford.edu/research/#zumarraga2026tshaystack): TS-Haystack benchmarks retrieval and reasoning across long time-series contexts and compares direct language-model approaches with tool-using retrieval.
- [Fine-Tuning LLMs in Behavioral Psychology for Scalable Health Coaching](https://schmiedmayerlab.stanford.edu/research/#mantena2025finetuning): This study examines whether fine-tuning on behavioral-science principles can support scalable, stage-aware physical-activity coaching.
- [LLMonFHIR: A Physician-Validated, Large Language Model-Based Mobile Application for Querying Patient Electronic Health Data](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2025llmonfhir): Physicians evaluate a mobile language-model interface for querying synthetic patient records, assessing response accuracy, understandability, and relevance. This pilot does not establish patient outcomes.
- [Dynamic Fog Computing for Enhanced LLM Execution in Medical Applications](https://schmiedmayerlab.stanford.edu/research/#zagar2025fog): This work explores dynamic fog-computing architectures for executing language-model workloads closer to sensitive medical data and end users.
- [Enhancing Distress Tolerance Skills in Adolescents With Anorexia Nervosa Through the BALANCE Mobile App: Feasibility and Acceptability Study](https://schmiedmayerlab.stanford.edu/research/#miranda2025balance): This feasibility study evaluates a mobile intervention designed to provide adolescents with timely distress-tolerance skills during high-stress situations.
- [GPTCoach: Towards LLM-Based Physical Activity Coaching](https://schmiedmayerlab.stanford.edu/research/#joerke2025gptcoach): GPTCoach uses a human-centered design process to study facilitative physical-activity coaching grounded in motivational interviewing and personal health data.
- [Impact of Workspace Environmental Features on Occupant Wellbeing: A Naturalistic Pilot Study](https://schmiedmayerlab.stanford.edu/research/#xu2025workspace): A naturalistic pilot study examining workspace environmental features and occupant wellbeing.
- [Medicine on the Edge: Comparative Performance Analysis of On-Device LLMs for Clinical Reasoning](https://schmiedmayerlab.stanford.edu/research/#nissen2025medicineedge): Medicine on the Edge benchmarks compact on-device language models for clinical reasoning across accuracy, efficiency, and device constraints.
- [Spatial Continuity: Investigating Use Cases of Spatial Computing for Users with Low Vision](https://schmiedmayerlab.stanford.edu/research/#heidekrueger2025spatial): Spatial Continuity explores how physical content can be transferred into an adaptable virtual environment for people with low vision.
- [Toward Scalable Access to Neurodevelopmental Screening: Insights, Implementation, and Challenges](https://schmiedmayerlab.stanford.edu/research/#bauer2025neurodevelopmental): This work describes an extensible platform for combining low-cost EEG collection with digital neurodevelopmental screening and identifies the validation and interoperability work still required.
- [Comprehensive Real Time Remote Monitoring for Parkinson's Disease Using Quantitative DigitoGraphy](https://schmiedmayerlab.stanford.edu/research/#hoffman2024parkinson): This work describes a connected platform for remotely measuring motor signs, tracking therapy context, and communicating results to care teams.
- [Utility of Smart Watches for Identifying Arrhythmias in Children](https://schmiedmayerlab.stanford.edu/research/#zahedivash2023smartwatches): This clinical study examines how consumer smart-watch recordings contributed to identifying arrhythmias in children evaluated at a pediatric center.
- [CardinalKit: open-source standards-based, interoperable mobile development platform to help translate the promise of digital health](https://schmiedmayerlab.stanford.edu/research/#aalami2023cardinalkit): CardinalKit describes an open-source mobile application and web-dashboard foundation for interoperable digital-health prototypes and research workflows.
- [Reducing the Impact of Breaking Changes to Web Service Clients During Web API Evolution](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2023breaking): This work presents a protocol-independent process for classifying web API changes and generating migration support for affected clients.
- [Enriching Cloud-Native Applications with Sustainability Features](https://schmiedmayerlab.stanford.edu/research/#vitali2023sustainability): This paper introduces an architecture for expressing sustainability features in cloud-native application design and carrying them into implementation.
- [The Path to a Modular and Standards-Based Digital Health Ecosystem](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2023ecosystem): This perspective argues for reusable, interoperable building blocks as the foundation of a sustainable open digital-health ecosystem.
- [Global Software Engineering in a Global Classroom](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2022global): This experience report distills lessons from a short, globally distributed software-engineering project spanning institutions, time zones, and connected devices.
- [Designing Evolvable Web Services](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2022dissertation): This dissertation develops methods and software abstractions for designing web services that can evolve across interfaces, deployments, and clients.
- [Transitioning to a Large-Scale Distributed Programming Course](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2020distributed): This paper reports how a large programming course moved to distributed delivery with remote infrastructure and pull-request-based feedback.
- [Apodini: An Internal Domain Specific Language to Design Web Services](https://schmiedmayerlab.stanford.edu/research/#schmiedmayer2020apodini): A domain-specific language approach to designing evolvable web services.
- [Fog Horizons – A Theoretical Concept to Enable Dynamic Fog Architectures](https://schmiedmayerlab.stanford.edu/research/#henze2019foghorizons): A theoretical model for dynamic fog-computing architectures.

## Resources

- [Plainly](https://github.com/SchmiedmayerLab/Plainly-iOS): An experimental app that helps patients ask questions about their own health records and prepare for clearer conversations with their care team.
- [My Heart Counts](https://myheartcounts.stanford.edu/): A longitudinal digital cardiovascular study connecting wearable sensing, patient-reported data, active tasks, health records, and personalized activity coaching.
- [Patient communication in reproductive endocrinology](https://schmiedmayerlab.stanford.edu/research/#trustworthy-agentic-systems): A clinical collaboration studying whether explanations grounded in health records can support reproductive health literacy and patient-provider communication.
- [Grove](https://github.com/SchmiedmayerLab/Grove): Our modular framework and ecosystem for building interoperable digital-health applications that can evolve with a study or care workflow.
- [SpineAI](https://schmiedmayerlab.stanford.edu/research/#trustworthy-agentic-systems): A research prototype exploring record-grounded explanations and evidence-based preparation for a spine-care consultation.
- [Patient needs in EHR messages](https://schmiedmayerlab.stanford.edu/research/#needs-figure-title): Extracting and clustering patient needs from portal conversations, then studying how they change across cardiovascular, kidney, and metabolic health.
- [OpenTSLM](https://www.opentslm.com/): An open family of models, code, and evaluation tools that makes time series a native modality for language-model reasoning.
- [STEWARD](https://arpa-h.gov/news-and-events/arpa-h-launches-worlds-first-bid-build-fda-authorized-clinical-ai-cardiovascular): Stanford’s supervisory system in the ARPA-H ADVOCATE program will screen actions proposed by cardiovascular care agents and give reviewers a rationale for flagged claims.
- [OpenMHC](https://neurips.cc/virtual/2026/poster/139369): A wearable-health dataset available to qualified researchers, with reproducible foundation-model baselines for prediction, imputation, and forecasting.
- [DIGITS](https://opendigits.dev/): An NIH-funded consortium making wearable and smartphone sensor data work across devices, platforms, and studies.
- [PocketTSLM](https://github.com/SchmiedmayerLab/PocketTSLM): On-device time-series language-model inference, bringing multimodal methods closer to the phones people already use.
- [TimeCap](https://github.com/SchmiedmayerLab/TimeCAP): A reproducible pipeline connecting wearable time series with language through deterministic, signal-grounded captions.
- [TimeNet](https://docs.timenet.ai/): A Python library and command-line interface for registering, fetching, and exploring heterogeneous time-series datasets through one shared format.
- [Heartwood](https://github.com/SchmiedmayerLab/heartwood): An open-source coding agent for biomedical research, connecting analysis work with reviewable actions, persistent sessions, and documented evidence.
- [Trove](https://github.com/SchmiedmayerLab/Trove): An open-source study platform in beta, connecting researcher configuration with participant enrollment, consent, and health-data collection.
- [one sec Stanford study](https://cdh.stanford.edu/adolescent-health-social-media-measuring-outcomes-digital-interventions): Now enrolling participants to study how brief pauses before social-media use relate to sleep, physical activity, and wellbeing, with a focus on adolescents.
- [HeartStrong](https://heartstronger.org/): A mobile-friendly, printable patient education resource for heart failure basics, goals, treatment conversations, and clinical visits.
- [ENGAGE-HF](https://github.com/SchmiedmayerLab/ENGAGE-HF-iOS): A family of connected-care research applications combining home measurements, patient-reported outcomes, education, and structured communication in heart failure.
- [NHLBI AI Data Science Center](https://nhlbi-ai-dsc.org/): A Stanford collaboration connecting secure research environments, multimodal models, and agent-assisted workflows for biomedical discovery.
- [Reproducible Manuscript Workflow](https://github.com/SchmiedmayerLab/PaperTemplate): Infrastructure for agent-assisted research and writing, keeping manuscript claims connected to references, analysis inputs, and reproducible figures.
- [Spatial interaction and accessibility](https://doi.org/10.1145/3663547.3759711): Earlier exploration of spatial interfaces for low-vision accessibility, with related interests in clinical education and surgical contexts.
- [TimeRLM](https://github.com/OpenTSLM/TimeRLM): A recursive language-model research artifact for precise, evidence-grounded anomaly localization in very long time series.

## Machine-readable records

- [Structured site index](https://schmiedmayerlab.stanford.edu/site-index.json)
- [Publication CSL-JSON](https://schmiedmayerlab.stanford.edu/publications.csl.json)
- [Publication BibTeX](https://schmiedmayerlab.stanford.edu/publications.bib)
- [XML sitemap](https://schmiedmayerlab.stanford.edu/sitemap-index.xml)