Migraine Forecast Architecture

🍑 Project Context

PeachyDay is a digital health startup helping patients manage chronic migraines. As a Data Scientist, I bridged the gap between raw data and business strategy by defining the product metrics framework and building the core ML features.

đź›  Technical Challenges

  1. Lack of Visibility: The team had raw data but lacked defined KPIs to measure product health or feature success.
  2. Clinical Validity: A “black box” model is useless in healthcare; predictions must align with medical understanding.
  3. User Retention: Users lacked motivation to log data daily without immediate feedback.

đź’ˇ My Solutions

1. Product Analytics & Metrics Framework

I established the internal analytics system to track product health and user behavior.

  • KPI Definition: Worked with stakeholders to define “North Star” metrics (e.g., DAU/MAU ratio, Feature Adoption Rate) and translated vague business questions into precise SQL queries.
  • Monitoring: Built automated dashboards to track user interaction flows, helping the team identify drop-off points in the onboarding process.
  • Impact: Transformed decision-making from intuition-based to data-driven, directly influencing the roadmap for the next 2 quarters.
PeachyDay Architecture

System Architecture with Physician-in-the-Loop

2. Physician-Guided Forecasting Engine

I developed an end-to-end prediction pipeline that translates clinical expertise into mathematical models.

  • Domain-Driven Feature Selection: Collaborated with clinicians to identify medically relevant triggers (e.g., focusing on barometric pressure changes) and filter out noisy variables.
  • Clinical Calibration: Applied post-processing techniques to smooth prediction curves based on doctors’ advice, preventing false alarms and modeling the gradual onset of attacks.
  • Model Performance: The final Random Forest model achieved an 18% accuracy boost over the baseline.

3. “Migraine Wrapped” Data Product

To improve retention, I engineered a Spotify-Wrapped style data story for 1,000+ users.

  • Visualization: Showcased longitudinal behavioral trends (e.g., identifying personal habits).
  • Impact: This personalized feedback loop drove a 12% increase in user engagement.

🚀 Key Results

  • Established the company’s first standardized Product Metrics Framework.
  • 18% Improvement in Forecast Accuracy via Clinical Feature Engineering.
  • 12% Uplift in User Engagement via Data Storytelling.
Visit PeachyDay Website

Note: Code is proprietary. This page demonstrates the methodology, clinical collaboration, and business impact.