Alan David Kumaran

Prism watches a fund portfolio the way an analyst team would if it never slept: it reads the documents managers send, keeps their claims comparable over time and flags drift, contradiction and concentration risk. Analysts ask questions in plain language and get answers that cite their sources. I designed and built the whole system: the data pipelines, the retrieval stack and the research agents.

I build the data infrastructure a modern asset manager runs on. I collect market data across custodians, reconcile it and automate the firm's reporting workflows. I also bring LLM tooling responsibly into a regulated financial-services environment.

Language models are good at guessing what might be true and bad at knowing whether it is, so this system splits the job. Agents form hypotheses about patterns in retail data and design experiments to test them. Statistical models do the validating: forecasting, anomaly detection and causal inference. People vet the results and the system rewrites its own prompts to produce more of what they accept.

The system ranks cryptocurrencies by momentum and bets on the leaders against the laggards, with equal money long and short. When the whole market moves, the two sides offset, so returns come from the gap between strong and weak coins. Position sizing adapts to volatility and the system rebalances itself on a fixed schedule.

Perpetual futures on different exchanges pay different funding rates. This system holds offsetting positions on both sides and collects the difference while staying market-neutral. Execution, hedging and risk limits run without me.