At Swissgrid AG, I designed a daily machine-learning pipeline for forecasting Swiss grid losses, informing procurement decisions and saving 10,000+ CHF per day. I introduced a Transformer-based architecture that improved forecasting accuracy by 15–20%.
For Redispatch Intelligence, I modeled and deployed three forecasting pipelines for the CH, IT, and FR markets, achieving 5% MAPE. I also evaluated an LLM-based grid-operator chatbot and deployed it with vLLM, FastAPI, and Pinecone, reaching under one second time to first token.
Previously, at Planetek Italia, I engineered MLOps workflows for computer vision projects, implemented a satellite-data indexing database with Redis caching, and architected an event-driven microservices layer using RabbitMQ. I also contribute to ThunderKittens, where I implemented CUDA kernels for Compressed Sparse Attention.
Italy