Francesco Forcher

Contributed to the design and development of a next-generation OLAP database for data warehousing: • Development of several engine and query language features, such as Boolean formulas and COUNT algorithms, with Rust. • Integration of AI/ML features into the product. Bringing Python and Julia scripts from research prototypes into production-ready containerized microservices, integrating state-of-the-art Gaussian Process based time series forecasting services optimized for low latency. • Worked on a custom Julia distributed logger for observability. • Research on a custom Kubernetes predictive autoscaler for our deployments. • Advanced custom load testing with Grafana k6. • Contributed to database API development, including gRPC and REST API. Implemented different caching and paging mechanisms.

Integration of a PostgreSQL database as data warehouse and its plugin Timescale for fast timeseries inserts and queries. • Used Talend and Airflow for ETL/ELT workloads. • Created interactive forecasts of demand using Facebook’s Prophet framework.

I developed advanced statistical routines using Python Pandas and scikit-learn to analyze crystal parameters for CERN’s UA9 experiment. • Enabled the analysis of very noisy datasets, with a significant impact in the tender process to award the crystal manufacture contract worth several hundred thousands CHF. • My thesis has been published as CERN internal note. • Using C++ ROOT library, I developed routines to analyze nuclear dechanneling for high energy particle beams in bent crystals, improved the simulation accuracy of SixTrack software.

Master's Degree in Computational Science and Engineering, Physics minor.

BSc in Physics.

Topic: "Effective High-Performance Computing and Data Analytics with GPUs"

ETHZ CSE Master Thesis at Paul Scherrer Institut, titled: “Intrusive Uncertainty Quantification of Maps” • Developed an application to perform intrusive Polynomial Chaos expansion to quantificate uncer- tainty in simulations based on approximate Hamiltonian maps, using symbolic computation. • A deep learning network approach has been evaluated to speed up the stochastic map calculation. • Developed in Python using SymPy and scaled into a distributed process using Dask. • Project: "Precise Simulation of a Compact μEDM Storage Ring". Utilizing scipy, a high-precision fully relativistic numerical simulation of a small scale muon storage ring has been developed.