We ingest and reconcile multiple independent sources: prices and market cap, derivatives (funding, open interest, liquidations, volume delta), on-chain and protocol activity (fees, revenue, TVL), and cross-asset data for equities and fixed income. Everything is consolidated into a unified feature store with one canonical identity per asset, and the universe is screened for liquidity and tradability so every candidate is executable.

A proprietary machine-learning model scores every asset by expected relative performance over a short horizon, learning from an in-house feature library across momentum, trend, volatility, volume, and statistical behaviour, normalized per-asset and cross-sectionally. The output is a systematic ranking that feeds portfolio construction.

Beyond price, we read the market's plumbing: derivatives positioning and funding, liquidation and volume-delta flows, protocol fees and revenue, network metrics, and market-wide indicators such as fear/greed and altcoin-season. This context helps the models separate durable moves from noise.

Alongside the ML model, a library of classical signals: momentum and trend-following, mean-reversion, volatility-regime detection, cross-asset relative strength, and macro breadth, combined through a weighted ensemble. Each is independently backtested and monitored, giving interpretable views next to the machine-learning output.

Portfolios are built with a two-stage robust Black-Litterman optimizer: a market-cap equilibrium prior, blended with our model- and signal-derived views, solved for weights that are robust to estimation error via an explicit uncertainty penalty. It respects position limits and full investment, and deliberately moves beyond classical Markowitz mean-variance.

The platform proposes; people decide. Every systematic allocation is reviewed by a weekly investment committee that owns risk, mandate compliance, and final sign-off. It brings the discipline of an asset manager to the speed of a research engine.