ML on Edge Hardware
Real-Time Energy Regression GNN on FPGA
A dynamic graph neural network for particle energy regression, deployed on an FPGA at 0.28 ms latency.
PyTorchGNNEdgeConvDGNNFlow
Coming soon
What I do
Skills & Tools
During my journey as research physicist I have been active in system operations, data analysis, ML/DL development and more; and filled positions that involved teaching, mentoring or project coordination. So I can proudly say I learned a lot!
Programming
Python, C++, SQL, Bash, git, AI-assisted development
Data & Analytics
NumPy · pandas · ROOT · distributed/batch computing (HTCondor, CRAB) · data-driven modeling · frequentist statistics · hypothesis testing
Machine learning & Ops
CI/CD · Docker · PyTorch · TensorFlow / (Q)Keras · scikit-learn · XGBoost · Grafana · Model compression (pruning, PTQ, QAT) · FPGA / ASIC co-design via HLS tools (hls4ml, conifer)
Soft skills
Collaboration · critical thinking · problem solving · communication · empathy · presentation · adaptability · time- and project-management · mentorship
Projects & experience
Selected work
ML on Edge Hardware
Electron/Photon Classification BDT on FPGA
The first ML algorithm for the CMS hardware-trigger upgrade: a BDT running in FPGA firmware at 100 ns latency. CMS Award 2023
XGBoostconiferBDTXilinx VU13P
Coming soon
Physics & Data Science
Searching for Supersymmetry & Dark Matter
Data-driven statistical modeling on petabyte-scale collision data to find what is Beyond the Standard Model of particle physics.
PythonROOTStatistical modelingLikelihood fits
Coming soon
ML Systems
HGCAL Data Compression & Clustering
On-detector autoencoder compression and clustering for a next-generation high-granularity calorimeter.
AutoencodersPyTorchEdge MLQuantization
Coming soon