The Unitarity Triangle, 2006–2023
Vincenzo Vagnoni’s 2024 LHCb Starterkit talk walked through two decades of CKM fits. Play the years: the bands shrink, they still meet, and LHCb Upgrade II is the next millimetre.
☕️About Me
I’m a postdoctoral researcher at Data Science for Digital Society (DS4DS), La Salle Campus Barcelona, working on AI for high-energy physics. I build lightweight graph neural networks for real-time particle reconstruction at the LHCb experiment and LLM/RAG systems that help physicists revive and accelerate their analyses. Before this, I spent 2.5 years at CERN operating and programming the CMS hadronic calorimeter, and earned my PhD at the University of Alabama searching for new physics in high-mass diphoton events. I care deeply about science communication: I’ve built museum exhibits, interactive physics visualizations, and open-source tools for healthcare and education.
Interests
- Real-time machine learning for particle physics
- LLM agents and retrieval for scientific analysis
- Fast inference and model compression
- Science communication and digital health
Education
PhD Physics (High Energy Physics)
University of Alabama
BSc Physics
Ateneo de Manila University
📚 Research
I work on making machine learning fast enough for physics that cannot wait. At the LHCb experiment, collisions happen 40 million times per second, so I design lightweight graph neural networks and knowledge-distilled models that reconstruct particles in real time within the trigger’s strict latency budget, targeting the next-generation PicoCal calorimeter for LHCb Upgrade II.
In parallel, I build LLM, RAG, and agentic systems for scientific analysis: retrieval over analysis code, notes, and documentation that lets physicists revive legacy measurements and accelerate new ones. I presented three talks on this research programme at CHEP 2026 in Bangkok.
Beyond the lab, I care about science communication and open-source tools for health and education, from museum exhibits to interactive physics visualizations you can explore on this site.
A year of new skills, reconstructed
Twelve months of my postdoc, drawn as an event display: each track is a skill area, each hit a milestone, each measurement a real outcome.
Real-time ML & model compression
8× faster inference
Attention-enhanced GarNet GNNs
Node-centric architectures for calorimetry
Knowledge distillation (~40% fewer parameters)
Graph→MLP distillation, ~95% smaller
Student beats teacher on energy resolution
Inference engineering
5× CPU · 2× GPU speedup
PyTorch → ONNX export pipelines
Multi-core CPU benchmarking
NVIDIA A100 GPU benchmarking
FP32 numerical parity at 10⁻⁷
Targeting the LHCb GPU trigger (HLT1/Allen)
LLM systems & agents
Legacy Λb→Λγ analysis revived
RAG over analysis scripts, notes & docs
Vector databases for physics knowledge
Agentic LLM workflows (Claude, GPT)
MCP servers for research tooling
Benchmarking vs non-RAG baselines
All three tracks were presented as talks at CHEP 2026 (Computing in High Energy and Nuclear Physics), Bangkok, May 2026.
Selected Publications
Full record on INSPIRE-HEP, ORCID, and GitHub.
(2024).
Search for new physics in high-mass diphoton events from proton-proton collisions at √s = 13 TeV.
JHEP 08 (2024) 215.
(2024).
Development of the CMS detector for the CERN LHC Run 3.
JINST 19 P05064.
(2018).
Search for physics beyond the standard model in high-mass diphoton events from proton-proton collisions at √s = 13 TeV.
Phys. Rev. D 98, 092001.
(2016).
Path Integral Solutions to the Distributions of Statistical Mechanics.
arXiv:1608.06280.
(2015).
Lorentz Dispersion Law from classical Hydrogen electron orbits in AC electric field via geometric algebra.
arXiv:1507.04509.
Recent Talks
Featured Projects
Recent News
Three talks at CHEP 2026 in Bangkok
Presented lightweight GNNs for LHCb’s PicoCal, PyTorch-to-ONNX inference optimization, and RAG-enhanced agentic revival of a legacy Λb → Λγ analysis at the 28th Conference on Computing in High Energy and Nuclear Physics.