Aimoto brings together data preparation, daily forecasting, backtesting and result comparison in a local research interface.
Overview
The problemEvaluating Bitcoin forecasting approaches requires preparing time series, repeating experiments and comparing results without mixing training and evaluation data.
The solutionA local application connects market data, configurable models, a backtesting engine and comparison views to support the complete experimental workflow.
Import or refresh daily series and select the features used in experiments. Feature catalogue, data providers and freshness indicators.
1 / 4: Prepare the data. Previous step / Next step
The interface calls a local API that orchestrates computations and persists experiment metadata and artifacts.
Next.js / TypeScript
Configure, track and compare experimentsFastAPI / Python
Validate requests and expose resultsPython
Run experiments and backtestsSQLite / Parquet
Store execution states, metrics and artifactsTime-series backtests can be biased if future observations leak into training.
Date-based walk-forward splits and explicit availability lags for features.
Model training and calculations can outlive HTTP requests and may be interrupted.
A job queue, separate worker, states and progress tracking, with SSE updates and polling fallback in the interface.
An interrupted write can leave incomplete result files.
Validated temporary JSON writes followed by atomic replacement, fsync and targeted retries for I/O errors.
Main technical stack
Run configuration flows and a batch matrix of feature and model combinations.
Forecasting and backtesting engines, run history, leaderboard and result visualizations.
A job queue, computation stages and status updates surfaced in the interface.