Founding Team
Built for the age of agentic finance: AI that understands cause and effect, explains every decision, and adapts in real time.
Where others see noise, we see information waiting to be resolved.
Every decision balances two goals simultaneously: what we gain drives the agent toward returns and goal attainment, while what we learn drives it toward reducing uncertainty before committing capital. Classic financial models only optimize the first term — and ignore the second entirely.
The Entropic Sharpe Ratio measures how efficiently a strategy converts belief updates into risk-adjusted returns. Where the traditional Sharpe Ratio captures return per unit of volatility, the ESR captures return per unit of informational work — connecting portfolio performance to the thermodynamics of decision-making. Strategies that generate high returns with minimal belief dissipation are informationally efficient.
Standard AI — from large language models to deep learning — is trained on historical data. When markets change, it fails silently. When it trades, it can't separate what the market did from what its own actions caused. And it can never explain why.
Transparent, auditable, and thermodynamically efficient — built for institutional trust.
Unified inference-control loop for single and multi-asset allocation. Regime-aware position sizing with built-in risk management. No separate risk engine required.
Full decision trace decomposition satisfies EU AI Act transparency requirements. Every trade decomposes into inspectable beliefs, expected free energy components, and confidence gates.
Entropy-based diagnostics replace backward-looking VaR. The Entropic Sharpe Ratio quantifies informational efficiency. Regime-conditional performance attribution is native to the framework.
Traditional finance — from Black-Scholes to DCF — is built on association: what has tended to happen. Fintropic evaluates policies under interventional distributions: what will happen if we trade. That distinction is the difference between a model that fails when regimes shift and one that holds.
We don't just build adaptive agents — we design them using autonomous scientific discovery. Leveraging SciAgents (Ghafarollahi & Buehler, Advanced Materials, 2025), multi-agent swarms explore vast knowledge graphs to uncover novel observation modalities, risk factors, and investment signals that human researchers would never think to combine.
Swarms of scientific agents traverse ontological knowledge graphs — connecting disparate domains from macroeconomics to supply chain dynamics to geopolitical signals — to identify novel investment factors with genuine predictive power.
Discovered factors become observation modalities in the agent's POMDP architecture. SciAgents configure the likelihood matrices, transition dynamics, and preference structures — automating the research process that traditionally requires teams of quants.
The designed agents are deployed with full transparency: every factor, every matrix, every belief traces back to an auditable discovery chain. From hypothesis to portfolio, the entire pipeline is autonomous, interpretable, and continuously evolving.
“Genuine discovery emerges from multi-agent interaction — adapting and co-creating through adversarial reasoning, shared memory, and in-situ learning.”
Whether you’re an asset manager, family office, or researcher — we’d love to hear from you.
