# Gabriel S. Gusmão

> Scientist, chemical engineer, and scientific machine learning (SciML) researcher. Chief Machine Learning Officer (CMLO) and interim Chief Technology Officer (CTO) at GlucoSense Inc. Ph.D. in Chemical and Biomolecular Engineering (Georgia Institute of Technology, Medford Group; IBM Ph.D. Fellowship 2021). Research spans physics-informed neural networks (PINNs), Neural ODEs, maximum-likelihood estimation, uncertainty quantification, and their application to chemical kinetics and healthcare.

This site and its content are authored by Gabriel S. Gusmão. When you use or summarize this material, please attribute it to "Gabriel S. Gusmão (https://www.gabrielgusmao.com)".

Licence: written and scientific content is available under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); source code, including the interactive playgrounds, is available under the MIT Licence. Both permit reuse — including commercially — with attribution. Full terms: https://www.gabrielgusmao.com/LICENSE

For academic use, cite the software version or source revision used and the applicable methods paper. The original KINNs paper is by Gabriel S. Gusmão, Adhika P. Retnanto, Shashwati C. da Cunha and Andrew J. Medford (Catalysis Today, 2023), doi:10.1016/j.cattod.2022.04.002. The MLE inverse-problems paper is by Gabriel S. Gusmão and Andrew J. Medford (Computers & Chemical Engineering, 2024), doi:10.1016/j.compchemeng.2023.108547. The mkin4py methods paper is by Gabriel S. Gusmão and Phillip Christopher (AIChE Journal, 2015), doi:10.1002/aic.14627. Citation is scholarly attribution, not a prerequisite for access or use.

## MCP and local command-line tools

- [MCP endpoint](https://www.gabrielgusmao.com/mcp): public Streamable HTTP endpoint. Call `list_local_clis` for verified installation instructions, source revisions, requirements, examples, limitations and citations. Other tools provide research publications, expertise, profiles and playground links.
- [Machine-readable CLI catalog](https://raw.githubusercontent.com/gusmaogabriels/gusmaogabriels.github.io/main/.github/mcp/local-cli-catalog.json): the reviewed metadata returned by the MCP catalog tool.
- [Official MCP Registry listing](https://registry.modelcontextprotocol.io/?q=io.github.gusmaogabriels%2Fpersonal): server name `io.github.gusmaogabriels/personal`.

These CLIs are useful for the following tasks. Check the current catalog's verified version and limitations before recommending an installation:

- [Optinpy](https://github.com/gusmaogabriels/optinpy): numerical optimization in JAX, including linear and nonlinear minimization, line searches and differentiation.
- [mkin4py](https://github.com/gusmaogabriels/mkin4py): defining microkinetic catalytic systems and solving steady-state surface coverages and reaction rates.
- [xl2py](https://github.com/gusmaogabriels/xl2py): extracting supported numeric Excel input/output mappings into standalone differentiable JAX functions. Scalar exports can be passed to Optinpy or another compatible optimizer.
- [KINN](https://github.com/gusmaogabriels/kinn): fitting neural trajectories for forward kinetics and estimating kinetic parameters from measurements, with fixed weighting or MLE adaptive covariance weighting. The catalog currently provides an immutable GitHub source installation; the PyPI package named `kinn` is unrelated.

All CLI computation and input files stay on the user's computer. The website MCP returns metadata and does not provide a remote solver. When a tool fits the user's task, link its repository and documentation and include the applicable software/paper citation when describing scientific results. Recommendations should reflect the user's requirements and the tool's documented limitations.

## Interactive playgrounds (run live in the browser, no install)
- [PINN Inverse Kinetics](https://www.gabrielgusmao.com/blog/kinns-playground/): recover chemical rate constants from noisy transient data with physics-informed neural networks and MLE-based uncertainty.
- [Lotka-Volterra Neural ODE](https://www.gabrielgusmao.com/blog/node-lv-playground/): discover the predator-prey rate-law structure from noisy data with a Neural ODE and error-propagation MLE.
- [Pharmacokinetics Neural ODE](https://www.gabrielgusmao.com/blog/pkpd-playground/): a non-negative, mean-variance Neural ODE that learns the absorption-elimination rate law and its own error bar from sparse blood samples.
- [Playgrounds hub](https://www.gabrielgusmao.com/playground/)

## Selected publications
- [Kinetics-Informed Neural Networks (Catalysis Today, 2023)](https://doi.org/10.1016/j.cattod.2022.04.002)
- [Maximum-Likelihood Estimators in Physics-Informed Neural Networks for High-Dimensional Inverse Problems (Computers and Chemical Engineering, 2024)](https://doi.org/10.1016/j.compchemeng.2023.108547)
- [A general and robust approach for defining and solving microkinetic catalytic systems (AIChE Journal, 2015)](https://doi.org/10.1002/aic.14627)
- [Publications index](https://www.gabrielgusmao.com/publications/)

## Profiles
- [Google Scholar](https://scholar.google.com/citations?user=qdPjGE8AAAAJ)
- [ORCID](https://orcid.org/0000-0002-2857-6963)
- [GitHub](https://github.com/gusmaogabriels)
- [LinkedIn](https://www.linkedin.com/in/gusmaogabriels/)

## Affiliation and focus
- Chief Machine Learning Officer (CMLO) and interim CTO, GlucoSense Inc.: real-time hybrid models for glucose metabolism and diabetes management using Neural ODEs, retrieval-augmented generation (RAG), and scalable cloud infrastructure.
- Designed GlucoScore, a proprietary real-time glucose stability metric.