Skip to main content

🧑‍🏫 Organizer Guide

AI-Guided SIR Modeling Hackathon

This document provides step-by-step guidance for organizers and facilitators running the AI-Guided SIR Modeling Hackathon.

It covers preparation, materials, facilitation flow, and post-hackathon follow-up to ensure a smooth and effective event.

🎯 Hackathon Goals (Organizer View)

The goals of this hackathon are to:

  • Teach core epidemic modeling concepts using a simple SIR framework
  • Demonstrate parameter inference from real outbreak data
  • Show how ChatGPT / LLMs can accelerate scientific coding and reasoning
  • Deliver a fully reproducible, beginner-friendly modeling workflow in 2 hours

This is a guided learning hackathon, not a competitive event.

⏱ Hackathon Format

  • Duration: 2 hours
  • Mode: Live guided demo + participant hands-on coding
  • Platform: Google Colab
  • Audience: Mixed background (data science, public health, general technical)

📦 Organizer Preparation Checklist

1️⃣ Google Colab Notebook (Required)

Prepare one master notebook that participants will copy.

Notebook requirements

  • Runs top-to-bottom without errors
  • Uses only standard scientific Python libraries
  • Includes # TODO: markers for participant interaction
  • Includes checkpoint plots after each major section

Recommended sections

  1. Title & objectives
  2. Dataset description (Niamey measles)
  3. Imports & configuration
  4. Data loading & exploratory plots
  5. Feature-based R₀ estimation
  6. SIR ODE model (biweekly time scale)
  7. Least squares fitting
  8. Poisson likelihood inference
  9. Profile likelihood / uncertainty
  10. Auto-generated Markdown summary

2️⃣ Dataset Preparation

Ensure the dataset is included in the repository:

  • File: data/niamey.csv
  • Columns:
    • biweek
    • community
    • measles

Important modeling notes to highlight

  • Data represent biweekly case counts
  • Infectious period ≈ 2 weeks
  • Time unit in models = biweeks
  • Source: Grais et al. (2006)

3️⃣ Pre-Recorded ChatGPT Demonstrations

To maintain pacing, prepare short screen recordings of ChatGPT interactions.

Recommended clips

  1. Notebook structure & assumptions
  2. SIR model implementation
  3. Feature-based R₀ estimation
  4. Incidence mapping (ΔH)
  5. Least squares vs Poisson inference
  6. Profile likelihood interpretation

Guidelines

  • Each clip: 30–90 seconds
  • Show the full prompt + ChatGPT response
  • Avoid scrolling or editing unrelated content

🧭 Facilitation Guide (During the Hackathon)

Key Emphases for Facilitators

  • Reinforce units (biweeks, not days)
  • Clarify what the data represent (counts vs states)
  • Encourage frequent plotting and sanity checks
  • Treat ChatGPT as an assistant, not an oracle

Time Management Tips

  • Prioritize:
    • SIR simulation
    • Feature-based R₀
    • One inference method (Least-squares or Likelihood)
  • Profile likelihood is optional if time is tight

⚠️ Common Pitfalls to Watch For

Issue How to Address
Confusing prevalence with incidence Re-emphasize ΔH mapping
Incorrect time units Repeat “biweeks” frequently
Over-trusting ChatGPT output Ask participants to validate plots
Overfitting early Start with community A only

📊 Expected Participant Outcomes

By the end of the hackathon, participants should have:

  • A working SIR model
  • At least one fitted parameter set
  • One observed vs fitted plot
  • A basic understanding of uncertainty

🧾 Post-Hackathon Actions

After the hackathon, organizers should:

  • Share the final notebook link
  • Publish slides (if used)
  • Invite participants to fork the repository
  • Optionally share extension challenges (SEIR, stochastic models)

📌 Reuse & Adaptation

This hackathon is designed to be reusable:

  • For short courses
  • For academic workshops
  • For internal training

Organizers are encouraged to fork and adapt the materials as needed.

📬 Questions or Issues

If issues arise during preparation or delivery, please raise them via:

  • GitHub Issues
  • Internal organizer communication channels

Thank you for helping deliver a clear, accessible, and impactful AI-guided modeling hackathon.