Model Card: Patrick Long
Healthcare AI • Neuroscience • Leadership
v2025.1
Leadership Style
Task Oriented
People Oriented
Introvert
Extrovert
Analyzer
Introverted, Task-oriented
Director
Extroverted, Task-oriented
Collaborator
Introverted, People-oriented
Promoter
Extroverted, People-oriented
Training Data
Academia
| PhD | Neuroscience, University of Vermont |
| Postdoc | Harvard Medical School, University of Michigan Medical School |
| Domain | Brain development, myelination, glioblastoma |
| Methods | Molecular biology, translational research |
Industry
| Current | Associate Director, AI Engineering @ IQVIA |
| Team | Global team (US, UK, India) |
| Domain | Real-world data (claims, EHR) |
| Therapeutics | Rare disease, oncology, metabolic, CNS |
| Prior | Tech transfer, biotech consulting |
Feature Importance
Scientific Thinking
0.30
First-principles reasoning, mechanistic understanding
Applied Healthcare ML
0.25
Predictive models on real-world clinical data
RWD Intuition
0.20
Understanding the data generating process
Leadership
0.15
Team building, cross-functional influence
Neuroscience Domain
0.10
Expertise in life science and translational research
Model Architecture Comparison
🔮 Transformer/LLM
Attends to context across multiple levels: individual, team, business, culture; draws on wide experience (bench → tech transfer → applied AI)
Diplomatic tendencies may soften directness; can be verbose
🌲 XGBoost
Iterative self-improvement; learns from mistakes; sees value in marginal gains
Builds intuition through practice rather than upfront; needs to see things play out
🕸️ GNN (Graph Neural Network)
Relational understanding; leverages connections between people, teams, stakeholders for better outcomes
Sensitive to network effects; consensus-building can slow decision-making
📈 Linear Regression
Reduces complexity for explainability; prioritizes actionable insights
Effectiveness varies by context (heteroscedastic); simplicity may mask overconfidence
Regularization
- L1 (Sparsity): Scientific training grounds me in mechanistic thinking, focusing on what's clinically meaningful vs. noise
- L2 (Smoothing): Working across clinical domains prevents over-specialization
- Dropout: Rotating across therapeutic areas, methods, and the pre-clinical to clinical spectrum
Known Limitations
- Benefits from experienced peers to escape local minima
- May require explicit negative feedback; implicit signals sometimes dropped
- Converges iteratively; flexibility can come at the cost of long-term directional certainty
- Context window limitations on administrative tasks
Best Deployed For
Building ML teams
Translating technical ↔ business
Clinical domain + ML integration
Patient identification models