SHARPELY AI

The firm builds signals primarily from alternative data sources — earnings call transcripts, central bank communications, supply-chain indicators, and satellite imagery — processed through transformer-based natural language models and traditional factor frameworks. The research pipeline is designed to generate generalizable features rather than dataset-specific overfits. Human researchers curate data partners and validate signal economic logic, but model architectures are selected programmatically based on out-of-sample performance.

The firm operates exclusively in liquid, electronically traded global equity and macro futures markets, covering North America, developed Europe, and liquid Asia-Pacific exchanges. Its mid-frequency strategy holds positions for days to weeks, sitting in the liquidity sweet spot between high-frequency market-making and long-only fundamental investing. Capacity is intentionally constrained to instrument universes where the firm's alternative data edge is statistically measurable.

Human intervention is limited to tail-risk circuit breakers — pre-defined volatility or correlation regimes that trigger position reduction — and operational oversight. The firm does not employ discretionary portfolio managers who adjust model output, a deliberate architectural choice that eliminates behavioral drag but also narrows the investable universe to markets where the models maintain persistent statistical advantage.

Investment decisions are driven by a rotating group of research leads whose models demonstrate the highest efficacy metrics during a given period. The firm does not operate with a traditional CIO hierarchy; portfolio construction and risk allocation are algorithmically determined, with research scientist accountability tied to signal decay rates and contribution-to-variance metrics rather than P&L attribution alone.

Most quantamental firms deploy AI as a screening or idea-generation layer that feeds a discretionary PM. SHARPELY AI runs an end-to-end automated pipeline — raw data to execution — without fundamental analysts checking the machine's work or portfolio managers hand-adjusting position sizes. This pure-systematic architecture eliminates behavioral conflicts but restricts the firm to data-dense, high-liquidity markets where model performance is robust.