timesfm

Zero-shot forecasting, tabular classification, and regression via MCP — exposes Google TimesFM 2.5 and TabFM v1.0.0 to AI assistants. Just attach a CSV and describe what you want to predict.

From-scratch reimplementation of Google TimesFM (time-series foundation model) plus the full pretraining pipeline Google never open-sourced. Trained and honestly evaluated at 70M.

A stock prediction application that uses Google's TimesFM (Time Series Foundation Model) to forecast stock prices from Yahoo Finance, with FastAPI serving as the backend API.

Tempolith: free, local-first time-series forecasting workbench. TimesFM and LightGBM side by side, backtests, diagnostics, anomaly detection, and what-if scenarios. Runs entirely on your machine.

Production-grade statistical arbitrage terminal using Google's TimesFM 2.5 and Kalman Filters for dynamic hedge ratio adaptation. Features a high-contrast Bloomberg-style dashboard, vectorized backtesting, and real-time news sentiment analysis.

Advanced Hybrid AI expert system for NASDAQ & Oil (WTI) ETF trading. Merges Quantitative ML, LLMs (Gemma 4, Gemini free or not), TimesFM 2.5, Visual Chart Analysis, and EIA Fundamentals for high-accuracy signals. Features dual-ticker strategy and Trading 212 execution.

Google's New TimesFM 2.5 time-series forecasting engine with automated financial market reports via Ollama, Gemma, and interactive Streamlit UI.

Self-hosted forecasting + prediction service. Five zero-shot time-series foundation models (Chronos-2, TimesFM 2.5, Moirai-2, Toto-1, Sundial) across six forecast types, plus nine supervised tabular ML backends (LightGBM, XGBoost, sklearn family) with calibrated / stacking / diversified meta-learners. Unified REST API + MCP server.

Time-Series-Forecast-Transformer working at the local container.

Zero-shot TSFM forecasts (Chronos-2, TimesFM) meet constrained Markowitz optimization for dynamic equity portfolios. ー S&P 500 forecasting & portfolio optimization.

Node.js/TypeScript reimplementation of Google Research's TimesFM 2.5 — a decoder-only foundation model for zero-shot time-series forecasting. Built on ONNX Runtime.

Benchmarking zero-shot and fine-tuned time series foundation models for process model forecasting on directly-follows time series from event logs.

In this project, I explore the use of Large Language Models (LLMs) for time series forecasting, focusing on the task of stock market prediction.

Walk-forward portfolio backtesting using TimesFM 2.5 quantile forecasting across US (S&P 100) and India (Nifty 50) equity markets. Uncertainty bands drive a mean-variance optimizer with return-shrinkage penalty.

Zero-shot foundation model based predictive autoscaler plugin for Kubernetes. Uses Google TimesFM to forecast workload and proactively scale pods.

Forecasting monthly sales using Google's TimesFM model.

Next year forecasts of Eurostat Economic and Tourism indicators per NUTS 2.

Embedded time-series anomaly detection engine. Dual-runtime (Node + Browser) support. 40+ forecasting models, RRCF streaming, TimesFM foundation model, Shapley attribution.

TimesFM stock chart forecasting app with AI future candles, volume, indicators, and replay mode.

Undergraduate thesis project: Intraday XAU/USD time-series forecasting comparing Macro-injected XGBoost against TimesFM Zero-Shot model.