Ask an epidemiological
question. Get a
validated simulation.
EpiChat is a conversational agent that turns a plain-language question into an agent-based simulation in Starsim, grounds it in published estimates and country data, and explains what it found, so decision-makers can explore scenarios themselves, without waiting for a modeler.
The configuration is what the app's own tools build for this question, run with the simulation service's code; the conversation and the reply are illustrative.
The bottleneck in computational epidemiology is expertise, not compute.
During COVID-19, demand for modeling vastly outpaced supply. WHO and Africa CDC flagged this as a critical pandemic-preparedness gap — especially in LMICs with limited domestic modeling capacity.
Policy officers, surveillance teams, and health analysts need fast, defensible answers — but the modelers who can give them are a narrow bottleneck.
surveillance officers
policy teams
local health ministries
The model handles language. Tools handle the simulation.
A conversation moves through six stages, shown under the chat as a step strip. At each one the agent calls a tool; the tool, not the model, writes the configuration, applies the data and runs the simulation, and its result appears as a card you can inspect.
A question, step by step.
- disease
- measles · SEIR database
- R₀
- 15 → β calibrated database, range 12–18
- periods
- latent 12 d · infectious 8 d, gamma database means, illustrative spread
- fatality
- 1.5% of infections database
- population
- Kenya's ages, births and deaths UN WPP 2024
- network
- age-structured Prem 2021 matrix for Kenya
- immune
- 88% at the start WHO/UNICEF first dose, 2024
- seeding
- 0.05% of agents default
- run
- 10,000 agents · 365 days · 5 runs
"In the median run of five, 10.3% of the population is infected within the year (9.7–10.9% across the five), with the peak on day 150. A campaign vaccinating 70% between days 89 and 119 brings the median run to 2.7%, but with it the five runs disagree widely (1.3–6.5%), and the app says so."
The run suggested four interventions and the app shows three as buttons; the first one's result is in the chart above. The conversation and the reply are illustrative.
It works end to end, and it shows its sources.
The web agent is in a pilot.
Take part.
EpiChat runs as a web app with sign-in. During the pilot it is open to study participants with an Emory email address, who agree at first sign-in that their conversations are recorded for the research. To take part from elsewhere, or to see a walkthrough, write to Yuke.
Where the data roadmap stands.
Data sources and implementation notes are in the docs.
Built at CIDMATH, Emory University.
Yuke Wang, PhD, MSPH
Yuke's research sits at the intersection of infectious disease epidemiology, environmental health, computational modeling, and AI. EpiChat is a prototype exploring whether LLMs can meaningfully widen access to epidemiological simulation.
Annie Wang
Annie is a master's student in the Epidemiology program at the Rollins School of Public Health. She earned her Bachelor's in Health Studies and Statistical Sciences from the University of Toronto. Her research interests focus on statistical analysis in public health, with experience in epidemiological research, data visualization, and applied health data science.
Questions you probably have.
Help us widen who gets
to do epidemiology.
Domain experts
Lend your eyes to benchmark questions and what a correct answer should look like for the diseases you work on.
COVID-19 · TB · cholera
Data contributors
Help find the cited values the database still lacks, so every scenario rests on published evidence rather than assumptions.
fatality by age · vaccine efficacy
Modelers for reference
Expand to other Epi modeling packages and provide validation support.
outbreak · intervention validation