Curiomode builds curiosity models: algorithms that direct an AI agent's limited attention toward what it can learn the most from. Natto, our first product, applies them to US equity trading.
Natto trades a rule-based strategy in US equities and keeps a sealed record of everything it does. Agents study those records and propose improvements to Natto, so the system keeps improving from its own evidence.
Screens a values-aligned stock universe and trades written, rule-based BUY and SELL logic.
Installs a fixed, verified release in the cloud, starts and stops cleanly, recovers from restarts, and seals the evidence of every session.
Agents review sessions, replay the rules on recorded market data, and test ideas in experiments registered before the results are known.
Each trade and each skipped trade is stored with the rule that fired, the market state around it, and when the data arrived, in a form both people and agents can read.
Sessions are sealed, checksummed, and independently restored before anything is deleted. A claim about what Natto did can be checked against its records.
Ideas are tested in preregistered experiments with protected holdout data, and failed experiments are kept so the next round learns from them.
Natto trades only companies that meet clear business and financial standards, and owns every share it buys outright, with its own cash. Simple rules, easy to explain, and easy to verify.
Curiomode is inspired by our founder's earlier research on how learners allocate attention. Radwan et al. measured curiosity-driven attention allocation in children exploring novel stimuli (Proceedings of the Cognitive Science Society, 2022).
Natto turns that question toward markets: which stocks, sessions, and hypotheses deserve attention next.
An agent that allocates its attention well learns faster and wastes less on noise.
Leads product and research, and builds Natto. Previously a Beckman Scholar at Emory and a research scientist at Stanford in the HAI program on curious, self-aware AI agents (inaugural Hoffman-Yee grant), working with MIT, UCSD, NYU, Daylight Design, and Kinedu on benchmarks, datasets and methods for studying how humans and machines learn. Trading equities since 2020.
Leads business and partnerships. Over 10 years building companies: co-founded Mystro and Source Wellness, and led the acquisition and sale of a restaurant corporation. Completed a Master's thesis in mindfulness and religious studies. Co-developed the trading strategy Natto runs.
We're building an API that serves Natto's labeled, as-of market context: market state, price and indicator context, labeled price moves, and data timing and quality.
We're looking for a small group of design partners, including values-based investing platforms, trading desks, and AI builders, to shape it with us.