Microsoft
IDEO
Essential Inc.

Microsoft

Industry Solution Engineering (ISE)

MICROSOFT

Reimagining ambulatory rooming and arrival through patient-centered, AI-enabled design

MICROSOFT

Designing an assistive AI in a variable, regulated care management workflow without breaking human accountability

MICROSOFT

Augmenting clinician risk assessment through AI‑enabled decision support

MICROSOFT

Accelerating learning and production for novice 3D artists through generative AI

MICROSOFT

Designing AI‑assisted antibiotic decision support that fits seamlessly into clinical workflows

Reimagining ambulatory rooming and arrival through patient-centered, AI-enabled design

MICROSOFT

THE PROBLEM

Ambulatory rooming workflows are high‑volume, time‑constrained, and cognitively demanding, placing significant documentation and coordination burden on care teams. Existing tools often obscure decision timing and workflow clarity, risking inefficiency while threatening clinician judgment, patient safety, and trust in outpatient settings.

SYSTEMS DESIGN

I led a multidisciplinary, participatory design engagement, working closely with clinicians, operational leaders, data science, and product partners to identify and shape AI‑enabled opportunities within ambulatory rooming workflows. Through field visits, journey mapping, and co‑design workshops, we synthesized clinical realities with technical constraints to define high‑value intervention points. The resulting system vision emphasized augmenting, not automating, clinical work, clarifying decision moments, reducing cognitive and documentation burden, and integrating AI responsibly into existing workflows. Insights were translated into a phased, responsible roadmap, balancing desirability, feasibility, viability, and safety.

IMPACT

The work aligned clinical, operational, and technical stakeholders around a shared vision for AI‑enabled ambulatory care and informed leadership investment decisions. The roadmap established a foundation for piloting and scaling responsible AI solutions across outpatient settings, with clear principles for trust, workflow fit, and clinical autonomy.

6 WEEKS | FACILITATION & DESIGN LEAD

Designing an assistive AI in a variable, regulated care management workflow without breaking human accountability

MICROSOFT

THE PROBLEM

Care managers supporting Medicaid populations must synthesize fragmented member information while completing lengthy health risk assessments under regulatory constraints. Manual documentation processes are time‑consuming and prone to omission, increasing cognitive burden and reducing time available for meaningful member interaction.

SYSTEMS DESIGN

I led a design effort in close collaboration with care managers, data scientists, product, and engineering to design an MVP to support assessment completion. The system consolidates historical member data and pre‑populates assessment responses for human review, enabling care managers to edit, accept, or discard AI‑generated content with appropriate transparency. Grounded in RAI principles, design decisions framed AI as decision support rather than decision replacement, protecting areas requiring human judgment. Front‑end interactions were intentionally connected to transcription quality, data readiness, and compliance expectations to support appropriate reliance and informed oversight.

IMPACT

Design choices encouraged vigilance and accountability while improving workflow efficiency. Following the MVP, experience patterns are now being replicated across additional assessment forms and expanded to other state programs, enabling scalable, human‑AI collaboration in Medicaid care management nationwide.

9 MONTHS | DESIGN LEAD, COACHED A JUNIOR DESIGNER FROM THE CLIENT SIDE

Augmenting clinician risk assessment through AI‑enabled decision support

MICROSOFT

The problem

Clinicians must assess blood clots risk using fragmented, unstructured patient data under severe time pressure. Existing tools added friction and were often bypassed, forcing physicians to rely on memory and heuristics, introducing variability, inaccuracy, and risks, especially in complex cases.

Systems design

Through a multidisciplinary, participatory design process, I partnered closely with physicians, data science, product, and engineering to design and deliver an AI‑assisted MVP that augments, not automates, clinical judgment. The system synthesizes and prioritizes patient‑specific risk signals, surfacing the most relevant context at the moment of decision‑making through reasonings and citations. Design choices intentionally preserved clinician autonomy while supporting guideline‑concordant care, ensuring the tool fit seamlessly into existing workflows and earned trust in high‑stakes clinical settings.

Impact

The MVP demonstrated value in reducing cognitive burden and improving clinician confidence, leading to scaling across the health system. The work established a repeatable model for responsible, human‑centered AI adoption in inpatient care.

6 MONTHS | DESIGN LEAD

Background

Venous Thromboembolism (VTE) is the formation of blood clots within the vascular system, which can result in deep vein thrombosis (DVT) in the legs or pulmonary embolism (PE) if the clot travels to the lungs, potentially causing severe respiratory complications or failure. In the U.S., approximately 900,000 people are affected by VTE annually, with 60,000–100,000 deaths each year, half of all cases associated with hospitalization, making timely and accurate risk assessment critical to preventing avoidable complications and improving patient outcomes.

At Johns Hopkins Medicine, clinicians relied on manual Epic workflows and validated but time‑intensive risk assessment tools, requiring extensive chart review across multiple screens. This created:

High cognitive and administrative burden for physicians
Inconsistent completion of risk assessments
Risk of missed or delayed prophylaxis decisions

Brief: Design an AI assistant tool that would inform clinicians' decisions in risk factor identification and assessment.

My contributions

Framed the right problem for a high‑stakes clinical AI system and champion a human-centered design approach throughout the project
Established a cross-disciplinary operating model that brought clinicians, client stakeholders, and internal teams together to make aligned, system-level decisions
Co-designed the end-to-end physician experience and product interactions
Generated human-centered success metrics beyond accuracy
Ensured safety, trust, and adoption in real clinical workflows

Grounding the work in clinical reality

Our approach to understanding the problem involved conducting exploratory research and engaging in field visits with physicians, allowing us to observe actual admission workflows, interruptions, and the pressures of time firsthand.

We carefully mapped the current methods of assessing VTE risk, examining the various tools, documentation, and mental shortcuts used.

From the very beginning, we developed concrete AI concepts to test our assumptions and provide clinicians with tangible solutions to respond to.

Insights

Physicians must search across vast amounts of unstructured EHR data, notes, discharge summaries, and historical records—where relevant VTE risk information is hard to find, inconsistently documented, and easy to misinterpret.
Clinical accuracy is not only a function of correct data, but of whether clinicians can find and trust the right information within severe time constraints.
Even a single 'yes' can be decisive in physicians' decision to administer or withhold VTE prophylaxis.
Physicians already have a concept of their patients, so if AI surprises them, they'll investigate further.

"The question we're trying to answer is, does this person need VTE prophylaxis or not? And is there any reason why I shouldn't give them VTE prophylaxis? The questions were phrased in such a way that for the risk factors, a yes means they need VTE prophylaxis. And a no means well, at least for that risk factor, they don't...if we have a whole bunch of NOs and one yes, they need it. And if we have all NOs, then they don't need prophylaxis. So, the YESes are important 'cause they're the ones that were gonna hinge our decision."

—Admitting Physician

Designing the way the team worked

Iterative experience prototyping and weekly working sessions across physicians and disciplines to co-design the key moments of experience

Live walkthroughs of real patient cases

Introduced early, imperfect concepts to make AI behavior tangible and accelerate shared understanding

Used analogous safety‑critical systems to reason about trust and failure

Decisions made in the open, with tradeoffs visible

Workstreams overview

The workstreams worked in parallel and built on each others' learnings.

Human & AI Experience Design drove requirements gathering which led to experimentation validating those requirements and generating hypothesis based on them. User design preferences were directly implemented in application.
Validated experiments were replicated in the orchestrator and surfaced in the application. Findings of experimentation inspired further discussion with the clinicians.
Application usage and feedback inspired further discussions with users, and iteration over requirements as well as testing further improvements in experimentation.

Designing for workflow, not screens

Co-designing and iterative experience prototyping of critical decision moments: AI suggestions for 20 risk indicators, citations, reasoning, failure modes, feedback loops, and onboarding experiences
Co-defining experiments and project success metrics
Mapping end‑to‑end clinical workflows, not just UI states or happy paths to see how the AI fits into the admission flow, interruptions, and time‑compressed decisions.
Making risk explicit and measurable: Failure modes, Over‑reliance risks, & feedback and correction loops

Human and AI experience experiments

How might we (HMW) create the most appropriate info architecture that meets the physician needs and accommodates the different data categories and related interaction patterns?
HMW define the solution success systematically, starting with what is desirable for the physicians?
HMW ensure a holistic approach where the solution is meeting the needs of different use cases and not only the happy path?
HMW co-create safeguards that prevent harm and support user control?
HMW empower users to review, contextualize, and edit the AI-generated content?
HMW intentionally design for the moments of failure?
HMW gather feedback from the physicians in a meaningful and practical way?

HMW gather feedback from the physicians in a meaningful and practical way?

Goals: user sentiments, where we made a mistake to improve the results, enrich the ground truth

Requirements: For the pilot it might be ok to add the feedback for all the risk indicators and citations but for a larger scale launch, feedback should be collected on a broad level rather than for each individual risk indicator, which simplifies the process:

"Give people a chance to add their open-ended feedback if they want to capture more detailed and valuable feedback, people be able to explain their ratings and provide specific suggestions for improvement." —Physician

HMW gather feedback from the physicians in a meaningful and practical way?

Championing a responsible approach

Fostering human-centered AI conversations: Relying on the HAX toolkit, Overreliance framework, and other RAI best practices to create materials to inspire cross-disciplinary collaboration

Setting the right foundation: Co-creating the impact assessment and other documents in response to ORA recommendations, and translating those to client-facing and design materials, including providing control for care managers

Bringing contextual knowledge through research: Uncovering care managers needs, concerns, and desires, and what they want to know about AI to inform use cases, disclaimer content, training materials, datasets, and evaluation criteria

From pilot to scale

The VTE AI Assistant was launched as a limited MVP, integrated into Epic, and validated with a small clinician cohort.

Validating the MVP in Real Clinical Context: Ran a live pilot with clinicians, returned to the field to observe real usage
Insights: Where the system supported judgment, Where friction or uncertainty remained, How clinicians actually used (or ignored) features

Following successful MVP validation, the solution became Use Case #1 of the Healthcare General Reasoner (HGR) at JHM. The architecture, UX patterns, and experimentation framework developed for VTE were explicitly reused and extended for subsequent clinical decision‑support initiatives, including Antimicrobial Stewardship (AMS).

Impact

Clinical workflow impact: Reduced manual chart review and improved access to risk indicators in one place, as validated through clinician feedback

Business impact:

First-of-its-kind CDS with GenAI at scale
110,000 admissions annually impacted
Enhanced care quality with strategic and financial impact
Enabled expansion beyond VTE into additional clinical domains using the same core UX and engineering foundation.

The JHM VTE project was selected for Engineering the Amazing, recognizing it as a standout example of co‑engineering, responsible AI, and real‑world clinical impact

Accelerating learning and production for novice 3D artists through generative AI

MICROSOFT

THE PROBLEM

Novice and transitioning 3D artists often experience steep learning curves when using professional creative tools such as Autodesk Maya. New users frequently interrupt their workflows to consult external tutorials or forums, slowing skill development and production progress. At the same time, early research revealed resistance to AI automation, participants expressed concern that handing over creative control could undermine artistic judgment, which they viewed as core to their professional value.

SYSTEMS DESIGN

Through interviews, diary studies, and co‑design workshops, the team identified evidence‑based opportunity areas where AI could responsibly assist while preserving artistic control over aesthetic outcomes. The resulting system concept leveraged natural‑language prompts to generate executable code within the primary workflow, enabling artists to complete complex or unfamiliar tasks without leaving the application. RAI considerations, impacting the learning pathways, were incorporated into failure modes, feedback loops, and transparency mechanisms.

IMPACT

Validation with Maya artists and Autodesk stakeholders informed opportunity selection and feature prioritization, shaping a responsible roadmap for integrating AI assistance into professional creative workflows. The work demonstrated how Human‑AI design can improve learning and efficiency without compromising creative control.

5 MONTHS | DESIGN LEAD

Designing AI‑assisted antibiotic decision support that fits seamlessly into clinical workflows

MICROSOFT

THE PROBLEM

Inappropriate antibiotic prescribing often occurs at critical decision points, such as ED‑to‑inpatient admission, where clinicians must rapidly interpret fragmented patient data alongside complex antimicrobial guidelines. While stewardship guidelines are available, they are frequently bypassed due to poor workflow fit, difficult navigation, and the cognitive burden of manually locating diagnosis‑relevant context across the HER.

SYSTEMS DESIGN

Led the design of an Epic‑aligned RAI stewardship experience by establishing a foundational information architecture composed of three primary data domains: AI Suggestions (e.g., agree/disagree with prior antibiotic choice), Clinical Guidelines (diagnosis and treatment pathways), Patient Data (decision‑relevant clinical context). Through clinician co‑design, structured diagnosis and treatment guidelines were paired with relevant patient data to create an initial information hierarchy supporting context‑aware prescribing decisions and the integration pathways were evaluated across Epic.

IMPACT

Foster clinical adoption by addressing workflow alignment, epic integration trust, and usability early in the process. Structuring guideline content alongside patient‑specific data enabled context‑aware decisions without overwhelming clinicians, while explainable AI recommendations improved confidence in suggested treatment changes.

TBD