Meet Pandemonium, an innovative COVID-19 risk assessment app and framework! Built by Quantum Risk Analytics, Inc., a charitable nonprofit founded by MIT alumni, it uses sophisticated machine learning algorithms to more accurately model the disease spread and provide a highly personalized evaluation of infection risks.

It’s not yet publicly launched as the developers want to add a few more major features and data sources before the official release. But the app is already in the testing phase that the team behind it encourages anyone to join.

You can learn more about Pandemonium and how it works, from a big interview that Richard Hamlin, CEO at Quantum Risk Analytics gave to the blog of AnyChart, the company whose JavaScript charting library is used to power all charts and maps in the app.

A few excerpts:

...

AnyChart: Where is this project now?

Richard Hamlin: Currently, users enter their details (such as where they have been and are planning to go, and the size and crowdedness of indoor spaces they visit, etc.) into a form on a web page. We use this information to compute the user’s accumulated risks of infection and fatal infection over a period of time. It is still in the development phase and we are testing our prototype app with outside test users in the US. (We welcome anyone interested in helping us test the app to email us at testing@pandemonium.dev.)

We have the foundation of our framework in place with its core functionality. This includes, among other features:

• integration of the micro-mechanistic transmission model(s) with the macro-epidemiological model
• modeling individuals alongside hierarchical geographic regions and demographic groups
• crucially, the ability to represent the arbitrary movement of people between regions through dynamic coupling
• integration of vaccination effectiveness and risk factor modeling.

...

AnyChart: It would be interesting to delve into the technical side of Pandemonium. Please describe the underlying technology.

Richard Hamlin: Currently the core technology underlying our risk assessment app and modeling framework is Bayesian inference via Pyro, a Python-based probabilistic programming language (PPL), but we plan to also use quantum machine learning (QML) in the future.

We build on the Pyro Epidemiology contribution package, a general stochastic compartmental inference and projection model framework. Most of the complexity needed to support all of our features is layered on top of this. It translates the data from the database and user into the flattened mathematical problem, runs/integrates underlying models, and translates back the results to correspond to the original hierarchical structures for returning/saving.

Our framework leverages object-oriented programming (OOP) and is very flexible and extensible, with interchangeable submodels and multiple types of coupling factors. The basic framework of the model allows for an arbitrary degree of demographic and spatial refinement and time-dependent interactions.

Micromodels, such as others’ aerosol airborne transmission models, are easily incorporated. We have implemented and incorporated MIT professor Martin Bazant’s well-mixed room model, based on “A guideline to limit indoor airborne transmission of COVID-19.”

Pandemonium supports vaccine effectiveness models of arbitrary complexity. These are isolated from other, dependent models, minimizing computational cost.

The ancillary mask compliance model uses the Tensorflow and Keras frameworks with convolutional neural networks and transfer learning to determine whether each person in a social media image is wearing a mask. This will be used for each geographic area to better estimate mask usage from publicly posted photos than survey data provides and thereby enable us to more accurately assess risk. The contextualizer will distinguish between candid and staged photos in order to reduce sampling biases.

Risk modeling adds another layer on top of the macro model with integrated submodels.

User schedule optimizations will be performed using a proxy objective function.

...

Read the complete interview HERE.