I build machine-learning and optimization tools for forecasting and quantitative finance

32 pull requests merged into libraries used across ML, quantum computing and finance

• OpenAI Tart: container-registry authentication with spaces between WWW-Authenticate directives, as RFC 7235 allows (#1377)
• NVIDIA cuOpt (2): nightly CI for its CVXPY integration on images without a C++ compiler (#2101), and a minimize default for the objective sense of its MPS data model (#2181)
• Apple MLX: reductions over very large arrays (32-bit index overflow on CPU and GPU) (#4549)
• Amazon Chronos: series lengths for categorical item ids in predict_df (#537)
• QuantLib (2): underflow in the non-central chi-square distribution (#2908), and a faster Heston variance mesher for small vol of vol (#2940)
• CVXPY (6; top external contributor by commits in the 90 days to October 22, 2026, and 34th of 239 contributors all time): silent wrong answers on mixed-integer problems with NVIDIA cuOpt (#3555), MIP warm start for SCIP (#3483), geo_mean on any shape (#3493), tuple axes in std, axis behavior of cvar, axis checks for matrix norms
• skfolio (6): VaR, EVaR and sample-weight fixes in the risk measures, drawdown contributions, the Bodnar-Okhrin shrinkage of expected returns, and sample weights in the empirical covariance (#417)
• NautilusTrader (2): wrong default moving averages in trading indicators
• QuantEcon: ARMA impulse responses and simulations that came out delayed when p > q (#976)
• Nixtla StatsForecast: finite-sample conformal prediction intervals, shared with the Conformal Seasonal Pool model (#1246)
• Unit8 Darts (2): conformal prediction intervals (#3227) and compatibility with pandas 3.1 (#3289)
• FinRL (AI4Finance): tradability check on the price instead of the first indicator in StockTradingEnv (#1456)
• Pasqal Pulser: lazy conjugate/negative tensors converted to NumPy without resolving them (#1130)
• gcsopt (graphs of convex sets): compatibility with the new CVXPY solving chain (#23)
• Also Qdrant, NumPyro (2) and Qiskit Experiments

In review: NVIDIA cuOpt (MPS writer), Microsoft FLAML, AutoGluon, Nixtla StatsForecast, Open GenAI Stack (formerly Llama Stack), MLX-VLM, SciPy.

• Research assistant to Prof. Semyon Malamud (Swiss Finance Institute, EPFL), replicating and extending large factor pricing models
• WorldQuant BRAIN: #2 in Italy, Gold Level, 19 submitted alphas
• Kaggle: silver medal, 76th of 4,186 teams, AI Agent Security (OpenAI × Google × IEEE)
• MQFS: open, reproducible quantitative-finance papers, e.g. market regimes from correlation structure, electricity risk for AI data centres, constraint-preserving QAOA on IBM quantum hardware
• Qiskit Advocate (IBM Quantum) and solver of Jane Street's monthly puzzles

Open to ML engineering, quant, software and data roles, including remote, part-time or project-based contracts, and to research projects. LinkedIn