timesfm3

Cadence: Error-Bounded Lossy Compression of Demand Time Series with a Time-Series Foundation Model (Google TimesFM 3)

A pure-MLX Swift port of Google Research's TimesFM-3, a 330M-parameter foundation model for time-series forecasting. It runs natively on Apple Silicon with no PyTorch dependency in the Swift package, and matches the companion Python/MLX port to ~1e-6.

Strumento per l’analisi e la previsione del SuperEnalotto che verifica l’indipendenza delle estrazioni, valida i metodi predittivi con backtesting walk-forward e confronta le previsioni di TimesFM 3.0 con il caso.

Privacy-preserving, locally-hosted multimodal AI compliance copilot for banking AML/SAR - RAG over regulations, zero-shot forecasting, and a QLoRA-tuned SAR writer. Retrieval Recall@4 0.955, anomaly F1 0.938.

Native .NET inference for the Kronos K-line and TimesFM 3.0 foundation model times series forecasting.

A native SwiftUI iOS app where a small language model orchestrates a time-series foundation model as a tool, entirely on the iPhone via Apple's MLX