Fleming Scientific was founded by John Heironimus in 2012 to develop a novel computational nonlinear dynamical approaches to modeling human biological and behavioral systems from passive time series observation. The goal was to develop a LEAN approach to utilizing passive time series data for description, classifcation and/or prediction of these systems.

By 2015, we had developed a field theory for extracting features that integrated chaos, synchrony and some aspects of fluid dynamics. The introduction of Tensor Flow that same year made it possible to fit extracted features to labels in a wide range of natural systems including animal health, internet traffic, derivatives trading, P&C losses, urban crime, cell expression, rotating machinery and biometric time series such as ECG.

By 2018, our ability to extract features had reached the point where neural networks were no longer useful. Instead, models could be fitted with simple rigorous processes.  Generally, all of our models since have been fitted with 0-3 degrees of freedom which reduces overfitting and other risks that are commonly assoicated with ML. They are all based on clear, stable, and very specific physics-based features which also stands is contrast to ML solutions based on purely mathematical or iterative processes. In extensive study, we have invariably found that our physics-based appproach is far more accurate and relaible under significantly different future conditions than convetional AI because it actually knows how to think about the problem in terms of the laws of nature. Not surprisngly, it also requires a lot less data.

In 2019, we began focusing on mobile phone captured auscultation data. Although long studied as acouctic data, auscultation data is actually time series observation of hemodynamical and pneumodynamical energy dissipation which  is "turbulence" and the physical manifestation of chaos. This view of turbulence is not to be confused with computational fluid dynamics that typically relies on numerical approaches such as Navier-Stokes and Reynolds to describe the disorderliness of fluid flow.

In our initial study of 15,000 ausculation recordings, we found that all were low dimensional (correlation dimension < 3) and chaotic (MLE > 0) regradless of patient, disease or organ of study. Consistent with this chaos-based view, we developed Time Series Dynamics Modeling or "TSD"  initally as phase space permutations on Takens' theorem. This theory holds that the dynamics of any low dimensional nonlinear dynamical system can be reconstructed from any (single) inifinitely sampled time series observation of that system.  Mobile  phones are attractive modalities because they can support recording speeds in excess of 200 khz and, of course, they offer about 7 billion distribution points for our products.

In 2020, we began clinical mHealth study of the potential of our approach to proxy common gold standards. Since that time, we have completed over 20 clinical studies in four different hopsital systems. We have showed that mobile phone captured auscultation data can predict findings from echocardiogram, right heart catheter, spirometry, MRI/CT, PCR and other gold standards and we have shown utility in chronic disease, infectious disease, trauma,  pregnancy and MCM. Our goal is not to replace gold standards but, rather, to make robust testing options available in situations where none exists today.

In conjunction with our research partners, we have amassed the world's most robust annotated data set of mobile phone captured auscultation recordings. Extensive simulation of the data suggests that extraction of features from some kind of state space is always more fruitful than from time series directly regardless of the processes used to create features. Whereas the literature focuses on specific processes such as MLE or cepstrum, our focus is on the optimization of state space. This approach is so robust that it obviates the need for neural networks entirely. We are immeasurably grateful for the support of teams at NIH and BARDA for their objectivity and support. Powerful stakeholders in the success of data and comptationally intensive AI solutions, such as big data and neural networks, have materially impacted our ability to access research funds.

In 2022, our focus on mHealth and related devices was organized separately under Tele-stethoscope Inc. (www.tele-stethoscope.com) and more information is available directly from that team. Their broadening focus includes other devices such as PPG, NIR and thermography. These noninvasive and inexpensive devices have exciting untapped potential in a chaos-based, time domain approach.

In 2025, we expanded TSD to include synthetic non-temporal series (SNS) analysis. Simulation of our previous studies suggest that it offers a more lightweight and accurate approach than temporal state space. Development of synthetic non-temporal series is currently our main research direction.