Expertise
What I work on
From generative backbone to developability prediction and bench handoff.
Generative Protein & Antibody Design
Designing antibody and nanobody (VHH) sequences with protein language models, discrete walk-jump diffusion, and energy-based models, optimized across multiple developability properties with Direct Preference Optimization and ranked by Pareto selection for wet-lab validation.
Foundation Models for Biologics
Architecting and pretraining transformer foundation models over antibody and protein sequence space, then fine-tuning them for downstream design and developability prediction, with distributed mixed-precision training and warm-start strategies.
Generative & LLM Systems
LLM applications, prompt engineering, retrieval-augmented generation, and parameter-efficient fine-tuning, including multi-task training methods open-sourced as MTL4AD.
MLOps & Production
Production machine learning on Databricks: reproducible pipelines, experiment tracking, CI/CD, model and data governance, and full artifact lineage.
Leadership & Technical Direction
Setting technical direction across system design, modeling, and MLOps, mentoring engineers, and partnering with clients to translate business challenges into technical strategy and roadmaps.
Research & Open Source
Selected work
Peer-reviewed publications and open-source scientific ML tools.
Language models can identify enzymatic binding sites in protein sequences
A transformer approach that locates enzymatic binding sites directly from protein sequence, improving prediction accuracy and reducing false positives versus prior baselines. Released open source as RXNAAMapper.
Integrating Genetic Algorithms and Language Models for Enhanced Enzyme Design
A language model assistant for biocatalysis
Accelerating material design with the Generative Toolkit for Scientific Discovery (GT4SD)
Biocatalysed synthesis planning using data-driven learning
Open source
Generative Toolkit for Scientific Discovery, extended for training and fine-tuning generative models.
End-to-end enzyme-optimization pipeline combining protein language models with genetic algorithms.
Transformer model for enzymatic binding-site prediction, with strong accuracy and false-positive gains over baseline.
Language-model assistant that automates common bioinformatics workflows.
Parameter-efficient fine-tuning for multi-task LLM training, improving adaptability and cross-domain knowledge integration.
Experience
Where I've worked
Research and production, from the lab bench to deployed pipelines.
Senior Machine Learning Engineer
May 2025 - PresentVisium SA · Lausanne, Switzerland
- Lead a biologics protein-optimization program for a global pharmaceutical company, owning candidate selection, scoring pipelines, and generative sequence-design methods (discrete walk-jump sampling, Direct Preference Optimization) to design optimized nanobody (VHH) sequences for wet-lab validation.
- Core technical driver on an antibody foundation-model initiative for a leading biotech, shaping the architecture: discrete walk-jump diffusion, a DiT-style denoiser warm-started from an antibody protein language model, an energy-based model with Langevin MCMC, Direct Preference Optimization, and Pareto selection.
- Drive ML best practices across system design, modeling, and MLOps, and mentor engineers across projects.
- Partner with clients to translate business challenges into technical roadmaps, authoring technical proposals and architecture decks alongside commercial and IP framing materials.
Python · PyTorch · Databricks · PySpark · diffusion & energy-based models · protein language models · MLOps
Pre-Doctoral Research Scientist
Jan 2022 - Mar 2025IBM Research · Zürich, Switzerland
- Developed parameter-efficient fine-tuning methods for multi-task LLM training, improving adaptability and cross-domain knowledge integration, open-sourced as MTL4AD.
- Built an end-to-end pipeline for enzyme optimization combining protein language models with genetic algorithms, open-sourced as Enzeptional.
- Engineered a transformer-based model for enzymatic binding-site prediction, improving accuracy 38% and reducing false positives 30% versus baseline, open-sourced as RXNAAMapper.
- Built a language-model assistant automating bioinformatics workflows, open-sourced as LM-ABC.
- Contributed to GT4SD, extending it for training and fine-tuning generative models for scientific discovery.
- Developed a molecular-dynamics framework to validate AI-generated protein designs, reducing laboratory failure rates.
PyTorch · Hugging Face · GT4SD · RDKit · GROMACS · multi-GPU training
Research Intern
Feb 2021 - Jul 2021IBM Research · Zürich, Switzerland
- Developed a synthesis-planning approach combining biocatalysis with transformer models to optimize synthetic pathways.
- Applied OpenNMT for transfer learning in chemical-reaction prediction, analyzing attention mechanisms to improve interpretability across reaction types.
PyTorch · OpenNMT · RDKit
Bioinformatics Project Lead
May 2020 - Sep 2020StemAway · California, USA (Remote)
- Led an international group of 30 students through all stages of gene-expression analysis.
- Built an automated QC pipeline in Bioconductor, cutting analysis time 50%.
R · Bioconductor · gene-expression analysis
Education
Jan 2022 - Mar 2025
Ph.D. in Biomedical Engineering
Eindhoven University of Technology
Research conducted at IBM Research, Zürich. Thesis: "Leveraging Large Language Models for Enzyme Design, Functional Modelling, and Optimization in Biocatalysis".
Sep 2019 - Oct 2021
M.Sc. in Data Science
University of Rome, La Sapienza
Rome, Italy.
Sep 2016 - Apr 2019
B.Sc. in Bioinformatics
ESCI, Pompeu Fabra University
Barcelona, Spain, including an exchange at University of Rome, La Sapienza.
About
I'm a Senior Machine Learning Engineer and researcher with over five years designing and shipping generative AI systems, from research to production. At Visium I set the technical direction on protein engineering problems, using large language, diffusion, and energy-based models to design optimized antibody and nanobody sequences that reach the bench.
My work spans research and delivery, from first-author publications to open-source tools and generative design pipelines running in production. I mentor engineers and partner directly with clients to turn business challenges into technical roadmaps. I am a native English, French, and Italian speaker, and fluent in Spanish.
Beyond the lab
Outside work, I referee football at local and regional levels in Italy, was an active member of the 6 AM Running Club in Zürich, and I am a keen traveler. The same instinct runs through all of it: finding structure in something that looks, at first, like noise.
Let's build something meaningful.
Open to research collaborations, foundation-model work, and conversations about generative protein design.
Contact me