Mathematician, physicist, and computational chemist in love with ab initio modeling, surface science, and catalysis.
I work on quantum-chemistry method development, bridging DFT, DFTB, force fields, and ML interatomic potentials.
I recently started my open-source journey and I take it seriously, building tools, sharing workflows, and keeping science transparent.

• Mixing neural networks with high-accuracy TB frameworks to bypass the U-dependent limitations of traditional DFT PBE
• Developing and testing MLIPs, making extensive use of MACE, UMA, CHGNet, M3GNet, for materials modeling
• Building reliable workflows for electronic-structure analysis, d-band descriptors, and high-throughput screenings

MACE • UMA • CHGNet • M3GNet • PyTorch • Active-learning pipelines • Δ-ML workflows

VASP • Quantum Espresso • DFTB+ • FHI-aims • GPAW • ASE

NumPy • SciPy • Pandas • Matplotlib • ASE automation • Data processing & visualization

MPI • Bash • Git • C++ • SLURM/Flux workflows • Reproducible computational environments

Creating and validating next-generation DFTB parameter sets supported by ML models, with a focus on accurate polaron localization and energetics.

Developed an orbital-resolved confinement-potential fitting scheme for DFTB, enabling fast and reliable descriptor-based high-throughput studies.

Directed training and validation of M3GNet and CHGNet models to understand solvation structure and dynamic behavior in complex environments.

Applied neural-network ensemble uncertainty estimates to improve interpretability of interfacial-dynamics datasets.

• Teaching and mentoring, because explaining things is the best way to learn them
• Quantum computing, learning how to correct qubits that refuse to behave
• Competitive Pokémon, a mix of strategy, probabilities, and questionable life choices

• Email: [email protected]
• LinkedIn: www.linkedin.com/in/filippo-balzaretti/

I’m at the early stages of contributing to open-source projects, but I’m committed to growing in this space and sharing tools, data, and workflows that make computational science more accessible and reproducible.