Maria Febus — Healthcare Data Scientist

Healthcare + AI Philosophy

How I think about building AI in healthcare.

Models in healthcare are not finished when they're accurate. They're finished when they can be trusted, explained, and used responsibly in real clinical and operational workflows. A few principles that guide how I approach this work.

Explainable AI is not optional

A model that can't tell a clinician why doesn't belong in a clinical decision. Interpretability isn't a feature — it's a precondition for trust, adoption, and safe deployment in healthcare.

Patient-centered, always

The patient is the end user, even when they're invisible from the model. Every architectural choice should be measured against whether it ultimately serves patient outcomes — not algorithmic elegance or organizational convenience.

Operational realism over benchmark theater

A model that wins on a test set but fails inside a clinical workflow has failed. Performance on paper means little if it can't be operationalized — monitored, maintained, and trusted by the people who use it daily.

AI augments judgment, never replaces it

The most valuable healthcare AI doesn't make decisions for clinicians and operators — it makes their decisions sharper. Decision support is a partnership between the model and the human expertise that contextualizes its output.

About

From project leadership to applied AI.

The value I bring to organizations is the ability to connect strategy with execution and turn complex challenges into results. I bring a customer- and business-focused mindset to translating healthcare needs into practical AI and data opportunities, navigating complexity and interdependencies, and driving disciplined execution across multiple functions, stakeholders, and priorities.

With a global mindset, I am comfortable working across organizational, functional, and geographic boundaries — bringing together diverse perspectives, building strong partnerships between business and technical teams, and aligning people around shared outcomes and measurable business value.

I also believe sustainable results come from strengthening the teams around the work. I foster collaboration, coaching, and mentoring, and help establish practical best practices and repeatable ways of working that improve execution beyond a single initiative.

My goal is not simply to deliver a solution, but to help organizations build the capabilities needed to turn AI and data opportunities into scalable, responsible solutions that deliver lasting business value.

I'm looking for hybrid roles that sit at the intersection of strategy, program/project management, AI product management, and hands-on data science — where I can shape strategy and roadmaps, translate healthcare needs into practical AI solutions, and still contribute technically. My goal is to bring together disciplined execution, product thinking, and applied data science to help healthcare organizations turn AI opportunities into real, sustainable outcomes.

15+

Years at J&J — supply chain & R&D

Data Science/AI Portfolio projects

2026

MS graduation · Aug

EN/ES

Bilingual

Selected Experience

Over fifteen years leading projects at Johnson & Johnson.

Over 15 years leading complex, cross-functional projects across Pharmaceutical Supply Chain and R&D at Johnson & Johnson. The examples below highlight several of the strategic initiatives that shaped my expertise in healthcare technology and innovation.

MedTech

MedTech ERP Portfolio Leadership

IT Senior Manager, MT IT SC ERP Portfolio Leader

Technology governance leadership for a $150M, two-year phase of a global, cross-workstream ERP transformation spanning Make and Order to Cash business capabilities, overseeing a portfolio of 100+ employees and contractors.

Problem

Executive leadership lacked holistic visibility into IT workstream financials — planned vs. actuals, estimate-to-complete, and resource allocation — across a global, multi-workstream ERP transformation program.

Approach

Led technology governance across a global, multi-workstream ERP transformation program spanning Make and Order to Cash capabilities, managing approximately $150M in technology investments over a two-year program phase, with a team of 100+ employees and contractors.

Impact

Delivered consolidated, executive-level visibility into workstream financials, resource allocation, and estimate-to-complete tracking — enabling more informed, data-driven decisions across the $150M program phase.

Portfolio Management Stakeholder Management and Reporting

R&D

Enterprise Data Lake Implementation

IT Project Manager

A centralized data platform unifying clinical operations and R&D data across the enterprise, replacing point-to-point interfaces and legacy reporting tools with standardized visualization tools.

Problem

Redundant interfaces and reporting tools left clinical ops and R&D data fragmented and inconsistently defined, slowing reliable data retrieval for strategic projects.

Approach

Drove implementation of a centralized Enterprise Data Lake, consolidating sources, harmonizing data across systems, and replacing redundant interfaces and legacy reporting tools with standardized visualization tools.

Impact

Established a single source of truth for clinical operations analytics, simplifying the data landscape and enabling strategic R&D initiatives that depended on harmonized, reliable data.

Data Strategy Data Governance Azure Analytics Enablement

R&D

PBMC Sample Journey

IT Project Manager

An automated sample tracking and traceability system replacing manual, CRO-fragmented Excel processes with centralized, real-time visibility.

Problem

PBMC (Peripheral Blood Mononuclear Cell) sample tracking relied on manual Excel-based processes across multiple CROs, making it difficult to trace sample journeys, maintain visibility across stages, and consolidate data into a single reliable view for stakeholders.

Approach

Led the team that built an automated data integration pipeline to consolidate CRO sample data into a central repository, eliminating manual Excel consolidation and enabling visualization-driven tracking.

Impact

Automated sample tracking, improving chain-of-custody traceability for audit-readiness and giving stakeholders self-service visibility into sample status and CRO performance. Recognized with the 2022 Innovation Leadership Award.

Data Traceability Cross-Functional Collaboration Scientific Data Systems Program Leadership

R&D

Governance Model Establishment

IT Project Manager

Designed and rolled out governance models across Clinical Operations IT — from portfolio-level investment oversight to program-level delivery frameworks.

Problem

Portfolio Governance: Senior IT leaders lacked visibility into the health of the $33M Clinical Operations IT portfolio.

iAWARE Governance: Integrated Analytics and Data Warehouse Program delivery lacked a repeatable process, with unclear roles and responsibilities.

Approach

Portfolio Governance: Led creation of a governance model with dashboards, project reviews, and early risk mitigation.

iAWARE Governance: Established clear roles and a repeatable stage-gate process spanning intake, estimation, onboarding, and delivery.

Impact

Gave senior leaders ongoing visibility into a $33M portfolio, while establishing a repeatable, accountable delivery framework for iAWARE program delivery.

Portfolio Governance Program Governance Risk Management Stage-Gate Process Design

Supply Chain

EPO/EPREX Bio-Manufacturing Simulation

IT Project Manager

Problem

Bio-manufacturing processes involve complex, interdependent constraints and resource considerations that are difficult to evaluate dynamically — making it challenging to run what-if scenarios and adapt plans quickly when conditions changed on the production floor.

Approach

Led the development of a simulation model of the EPO bio-manufacturing process to identify and evaluate key constraints and resource bottlenecks, enabling dynamic what-if scenario analysis and generating updated manufacturing schedules with estimated timelines.

Impact

Identified a recommended schedule and vial-opening frequency change with the potential to enable one additional batch, plus optimal preventive maintenance windows — an estimated multi-million-dollar opportunity. Recognized with the Johnson & Johnson IMAGE Award for innovation in simulation-based manufacturing analysis.

Discrete Event Simulation Manufacturing Operations

Supply Chain

NPMonitor — End-to-End Product Launch Visibility

IT Project Manager / Scrum Master

A reporting portal delivering end-to-end product launch visibility, from drug substance through drug product manufacturing.

Problem

Senior leaders lacked a unified view into product status across the development and launch lifecycle, relying on fragmented, manual status updates across drug substance and drug product manufacturing stages — slowing early risk identification and decision-making.

Approach

Serving in a dual IT Project Manager/Scrum Master role, drove implementation of the NPMonitor reporting portal, replacing fragmented manual updates with a single cross-functional view of product status from drug substance through drug product manufacturing.

Impact

Gave senior leaders end-to-end visibility into product launch status, accelerating decision-making and improving early risk identification across the development and launch lifecycle.

Agile/SCRUM Delivery Data Analytics Manufacturing Operations

Selected Work

A governance framework and five projects spanning epidemiology, medical imaging, cardiovascular risk prediction, supply chain, and patient behavior.

AI Governance · Solutions Delivery

In Progress August 2026

AI Solutions Delivery Playbook — Beyond the Model

A seven-phase framework for delivering secure, responsible, and compliant AI solutions, from idea to impact. Target Audience: Cross-functional AI delivery teams — product managers, data scientists, ML engineers, and compliance partners — working in regulated environments such as healthcare and pharma.

Problem

Responsible AI guidance stops at principles, leaving regulated teams to invent controls mid-build — inconsistent evidence, late compliance surprises, and models that stall before production.

Approach

A delivery lifecycle from Define through Hypercare, with governance embedded at each gate: RACI ownership, required artifacts, validation criteria, and traceability across 17 regulatory sources.

Impact

A repeatable path to production that generates audit-ready evidence as a byproduct of delivery, not a scramble at the end.

AI Governance Responsible AI CPMAI RACI & Decision Gates Model Validation Regulatory Mapping MLOps

Read the Playbook

Epidemiology · Public Health

Completed February 2026

Diabetes Prevalence & Chronic Disease Correlation

A cross-state analysis of diabetes prevalence and its correlation with other chronic conditions across all 50 US states. Target Audience: Public health agencies, healthcare policymakers, community health organizations, and population health analysts supporting disease prevention initiatives and resource allocation.

Problem

Population-level chronic disease patterns are buried in state-level public health data, making it hard to target prevention strategy and program design.

Approach

Cross-state correlation analysis combining diabetes prevalence with comorbidity indicators, surfaced through statistical analysis and data visualization.

Impact

Patterns that inform payer risk modeling, chronic care program design, and population health prioritization.

Python Pandas Statistical Analysis Data Visualization

View on GitHub View Diabetes Prevelance UX Mockup

Deep Learning · Medical Imaging

Completed March 2026

Breast Cancer Detection — CNN Classifier

A convolutional neural network for classifying breast histopathology images as benign or malignant, with a clinical focus on maximizing recall to minimize missed diagnoses. Target Audience: Pathologists, oncology teams, and healthcare organizations exploring AI-assisted diagnostic tools to improve cancer detection accuracy and support clinical decision-making.

Problem

Manual histopathology review is time-consuming, subjective, and dependent on specialist availability — creating bottlenecks in early breast cancer diagnosis.

Approach

Custom CNN trained on the IDC histopathology dataset (224×224px patches), optimized for recall on malignant class — because in oncology, a missed malignant case carries far greater cost than a false alarm.

Impact

A scalable AI-assisted triage tool that flags high-risk tissue patches for clinician review, supporting faster and more consistent diagnostic workflows.

TensorFlow Keras CNN Medical Imaging scikit-learn

View on GitHub Watch Demo Video

Classification · Clustering · Cardiology

CompletedJune 2026

Predicting Myocardial Infarction and Identifying High-Risk Profiles

Phase 1: Supervised classification combined with unsupervised clustering to identify patients at risk of myocardial infarction and surface distinct high-risk profiles, enabling earlier, more targeted clinical intervention.Target Audience: Clinicians and care coordinators seeking to identify high-risk patients earlier and prioritize preventive interventions.
Phase 2 (Independent extension): Agentic AI system where specialized agents collaborate to interpret patient risk, retrieve clinical evidence, and support real-time clinical decisions. This phase moves the project from prediction to actionable decision support. A prototype of the CardioAssist clinician interface is available on GitHub.

Problem

Heart attacks remain a leading cause of death in the US. Traditional risk models produce a single score but rarely explain which patient profiles drive elevated risk — leaving population health strategies with prediction but limited direction for targeted intervention.

Approach

Supervised classification (Logistic Regression, Random Forest, XGBoost) with SHAP-based feature importance, ablation, sensitivity, and failure mode analysis — combined with unsupervised clustering (K-Prototypes, GMM) using the CDC's Behavioral Risk Factor Surveillance System (2024).

Impact

An interpretable risk framework that surfaces distinct high-risk profiles — informing targeted screening, patient outreach, and prevention-focused population health strategies.

Logistic Regression Random Forest XGBoost SHAP K-Prototypes GMM

View on GitHub View CardioAssist UX Mockup

CLASSIFICATION · SUPPLY CHAIN · PHARMA OPERATIONS

Expected July 2026

Predicting Drug Shortages Before They Happen

A classification model that flags pharmaceutical products at elevated risk of supply shortage — a real industry problem I've watched unfold from the inside. Target Audience: Pharmaceutical manufacturers, hospital pharmacy leaders, healthcare systems, distributors, and government agencies responsible for maintaining drug availability and strengthening supply chain resilience.

Problem

Drug shortages disrupt patient care and cost the healthcare system billions, yet they're often identified only after they've begun — too late for proactive mitigation.

Approach

Classification model trained on FDA shortage data, with features engineered from manufacturer concentration, drug class, sterile-injectable flags, and supply chain signals.

Impact

Early-warning capability that supports proactive inventory planning, alternative sourcing, and continuity of patient access through domain-informed feature engineering based on experience supporting pharmaceutical manufacturing and supply chain operations. Target Audience: Health plans, pharmaceutical patient support programs, providers, care management organizations, and value-based care teams seeking to personalize patient engagement and improve medication adherence.

XGBoost FDA Open Data Feature Engineering SHAP

Coming July 2026 View Drug Shortage UX Mockup

CLASSIFICATION · HEALTHCARE · PATIENT OUTCOMES

Capstone Expected August 2026

Medication Adherence Segmentation

A combined supervised and unsupervised learning approach to segment patients by medication adherence patterns. Target Audience: Health plans, pharmaceutical patient support programs, providers, care management organizations, and value-based care teams seeking to personalize patient engagement and improve medication adherence.

Problem

Non-adherence costs the US healthcare system over $300B annually, yet most interventions treat all non-adherent patients alike — missing the distinct behavioral patterns that drive different solutions.

Approach

Combined unsupervised clustering to discover adherence archetypes, then supervised classification to assign new patients to the segment most likely to predict their behavior.

Impact

A foundation for tailored adherence interventions — informing pharmacy programs, patient support design, and real-world evidence studies with behavioral nuance most models miss.

Clustering Classification Feature Engineering scikit-learn

Coming August 2026 View Medication Adherence UX Mockup

Bridging healthcare experience with applied data science & AI.

How I think about building AI in healthcare.

Explainable AI is not optional

Patient-centered, always

Operational realism over benchmark theater

AI augments judgment, never replaces it

From project leadership to applied AI.

Over fifteen years leading projects at Johnson & Johnson.

MedTech ERP Portfolio Leadership

Enterprise Data Lake Implementation

PBMC Sample Journey

Governance Model Establishment

EPO/EPREX Bio-Manufacturing Simulation

NPMonitor — End-to-End Product Launch Visibility

A governance framework and five projects spanning epidemiology, medical imaging, cardiovascular risk prediction, supply chain, and patient behavior.

AI Solutions Delivery Playbook — Beyond the Model

Diabetes Prevalence & Chronic Disease Correlation

Breast Cancer Detection — CNN Classifier

Predicting Myocardial Infarction and Identifying High-Risk Profiles

Predicting Drug Shortages Before They Happen

Medication Adherence Segmentation

Education and certifications, in progress and complete.

M.S. Applied Data Science and AI

MIT — AI in Healthcare

MIT — Applied Data Science

Agentic AI Professional Certificate

CPMAI Certification

Project & Agile Certifications

Let's talk about healthcare AI.