Intraday Trading · Setup Detection

Your personal
trading agent.

Arthira is an AI-powered pipeline for intraday setup detection. Label your own setups, train a lightweight classifier, and evaluate predictions — all in a tight feedback loop.

Capabilities

Everything you need. Nothing you don't.

A tightly scoped system built around one task: finding high-quality trading setups.

Setup Detection

Classify every bar as Long, Short, or None using a compact classifier trained on your own hand-labeled examples.

Interactive Labeler

Browser-based charting tool with multi-timeframe views, RSI pane, keyboard shortcuts, and per-theme settings. Label at speed.

Iterative Training

Auto-versioned checkpoints (v1, v2, …). Each round: label more setups → retrain → visualise predictions → repeat.

Focused Feature Set

A curated set of momentum and structure features — RSI dynamics, EMA channel distances, volatility normalisation. Signal without noise.

Class Balance Control

Configurable negative sampling keeps non-setup bars manageable without swamping real setups. Class-weighted loss handles the rest.

Lightweight by Design

Trains in minutes on CPU. Inference in microseconds. No GPU required — anywhere in the pipeline.

Workflow

How it works

A four-step loop that converges on a model that recognises your setups.

01

Label setups

Open the browser-based charting tool. Browse historical bars across multiple timeframes and mark each setup as Long, Short, or skip it. Your labels become the training signal.

Browser labeler · multi-timeframe
02

Train the model

Kick off training. The classifier learns from your labels, automatically balances the class distribution, fits a feature scaler, and saves a versioned checkpoint.

Auto-versioned checkpoints
03

Evaluate predictions

Run the evaluator on a held-out date range. An interactive chart overlays the model's predicted triangles on OHLCV bars alongside your ground-truth labels.

Interactive Plotly HTML output
04

Refine and repeat

Study the missed setups and false positives. Add the hard examples back into the labeler and retrain. Each round the model gets sharper.

Iterative until satisfied