Concentrated conviction on assets and margin

This Sophon-3 AI model flow starts with the DIJA-30 as the asset universe, then chains together three AI model nodes: Sophon-3 Screener, Sophon-3 Forecast, and a second Sophon-3 Screener before portfolio construction.

The idea is simple: first narrow the universe to more interesting candidates, then forecast their near-term return potential, then select the Top N forecasts to optimize against a benchmark before sending the results into portfolio construction. This creates a practical workflow where screening, forecasting, selection, and position sizing work together instead of as separate steps.

In this template, the first Sophon-3 Screener uses data types including Total Current Assets and Operating Margin. Total Current Assets is the sum of assets the company expects to convert to cash or use within roughly the next year (for example, cash, receivables, and inventory), which helps describe near-term resources and operating flexibility. Operating Margin measures operating profit as a percentage of revenue and helps show how efficiently the company’s core business turns sales into operating earnings.

Together, these inputs give the screener both a near-term resources lens (what the company has available soon) and an operating efficiency lens (how much profit it generates from sales before financing and taxes).

Sophon-3 Screener is an AI model node that continually learns from new market dynamics instead of relying on fixed screening rules. Its role is to filter the DIJA-30 toward stocks with stronger potential versus the benchmark.

The selected names are then passed to Sophon-3 Forecast, another AI model node, built for continual deep learning forecasting. Sophon-3 Forecast focuses more directly on price dynamics and estimates near-term return direction and magnitude.

After forecasting, the second Sophon-3 Screener ranks and selects the Top N forecasts to create a tighter candidate set for benchmark-aware optimization in portfolio construction.

Chaining the nodes is meaningful because they solve different parts of the problem. The first Sophon-3 Screener reduces noise by finding better candidates, Sophon-3 Forecast adds timing by estimating near-term return potential, and the second Sophon-3 Screener concentrates the signal by selecting the strongest forecasts before optimization and position sizing.