How Aaru uses Exa Snapshot to predict the future

How Aaru uses Exa Snapshot to predict the future

Did your agent predict the future, or just look up the answer?

Aaru builds simulations of human behavior to predict world events at scale. Their agents need recent, comprehensive context from across the web to reflect how people think today.

But to prove a simulation works, Aaru has to backtest it, so each agent can only see the web as it existed before the event it’s predicting. In their latest research, Exa was the only external retrieval tool they used.

In this 30-minute live session, Shyam Kumar (Head of Product, Exa) and Shashank Vemuri (Research Engineer, Aaru) walk through how Aaru builds and scores populations of simulated agents, and how Exa Snapshot keeps their backtests free of leakage.

What you'll learn:

• How to run leakage-free backtests by limiting web retrieval to what existed before a given date
• How versioned search (unknown URLs) and versioned contents (known URLs) works
• How Aaru builds demographically accurate agent populations and scores them against real survey data

Who should attend:

• Teams building simulations, synthetic respondents or agent-based models of human behavior
• Market research and insights teams comparing synthetic data with traditional surveys
• Forecasting, quant and policy teams backtesting models that use web data
• ML and evals engineers who need reproducible, point-in-time web context