Darius A. Faroughy, Ph.D.
Download PDF ↓I build generative AI for scientific discovery. Trained as a theoretical particle physicist, I now research at the intersection of generative modeling and the natural sciences, with a focus on flow-matching, diffusion, foundation models, and agentic systems for complex scientific data. Recent contributions include introducing the first multimodal generative model for collider data, developing the first foundation model trained on real data from the Large Hadron Collider (LHC), and demonstrating neural scaling laws for jet generation. Beyond physics, I led the development of a generative foundation model for pharmacokinetics time-series.
- Flow-matching models for large-scale structure of the Universe and other cosmological data.
- LLM-guided evolutionary search for de novo design of synthesizable molecules.
- Autonomous agents for scientific reasoning and discovery.
Technical Skills
- ML: Generative AI, Transformers, Flow-Matching, Diffusion, Neural Operators, Time-series, LLMs, Agentic AI
- Languages & Frameworks: Python, PyTorch, Lightning, HuggingFace, scikit-learn, NumPy, C++
- Computing: Bash, Git, Distributed GPU, SLURM, Comet
Experience
Postdoctoral Researcher2022 – Present
Rutgers University · New High Energy Theory Center, NJ, USA
- My Chemical Harness. Built the agentic harness and ran the molecular-design experiments for My Chemical Harness, an LLM-guided evolutionary framework over executable synthetic routes. On a soluble epoxide hydrolase proxy task, our agentic algorithm reached state-of-the-art performance across multiple molecular-design metrics — including sEH score, synthetic accessibility, and AiZynthFinder success rate — surpassing dedicated synthesis-aware generative baselines such as NVIDIA's ReaSyn.
- Foundation model for Pharmacokinetics. Led a multidisciplinary collaboration with mathematics and pharmacology groups to develop a generative foundation model for pharmacokinetic time-series, enabling zero-shot population synthesis and individual drug-response forecasting. Our method outperforms prior generative approaches.
- Collider-Bench. Led the design and release of an agentic-AI benchmark grounded in real scientific tasks, evaluating autonomous reproduction of experimental particle-physics analyses at the LHC.
- Multimodal flow-matching for particle clouds. Built the first multimodal generative model for point-cloud data from hadron colliders, jointly modeling continuous kinematics with flow-matching and discrete particle attributes (charge, flavor) via Markov jump processes. The resulting transformer-based model achieves current state-of-the-art.
- Foundation model for Particle Physics — Aspen Open Jets. Led the development of the first foundation model trained on real Large Hadron Collider (LHC) data, alongside AspenOpenJets — the first large-scale real-world LHC dataset for ML.
- Neural scaling laws for jet generation. Co-led the first systematic study of neural scaling laws for generative models trained on real LHC data, establishing predictable performance gains with model size.
Postdoctoral Researcher2019 – 2022
University of Zürich · Physik-Institut, Zürich, Switzerland
- NLP and Bayesian models for anomaly detection. Applied natural-language-processing techniques to tokenized jet substructure and Bayesian probabilistic modeling to anomaly detection at colliders. Contributed to the LHC Olympics 2020 community anomaly-detection challenge.
- Effective field theories. Classified flavor operators in the Standard Model Effective Field Theory under various flavor symmetries, and developed methods combining low- and high-energy constraints on flavorful new physics.
- Flavor physics. Investigated the interplay between low-energy flavor observables and high-momentum LHC searches, applied to leptoquark and Pati-Salam-like UV-complete models proposed as solutions to the B-physics anomalies.
- HighPT. Led the development of HighPT, an open-source Mathematica tool for global analyses of high-momentum Drell-Yan scattering within the Standard Model Effective Field Theory and beyond. Widely adopted by the community for new-physics searches at the LHC.
Teaching Assistant2019 – 2022
ETH Zürich & University of Zürich, Switzerland
- Teaching assistant for 6 semesters for master's-level courses in Quantum Field Theory I & II, Particle Physics, and the Standard Model of Particle Physics.
Education
- Ph.D. in Theoretical Particle Physics 2014 – 2019
Jožef Stefan Institute, Ljubljana, Slovenia - Postgraduate Diploma in High Energy Physics 2011 – 2013
ICTP, Trieste, Italy - B.Sc. in Physics 2005 – 2010
Universidad Simón Bolívar, Caracas, Venezuela
Full list of publications → Selected Publications · Google Scholar