TechPrufer
Get started
Grab
Grab logoGrab · Feature trail

AI-led Research Simulation Platform

1 post · Oct 2026

What this trail covers

Grab has developed sim-rs, a self-contained simulator for marketplace operations, enabling rapid experimentation with dispatch strategies. This platform connects to AI agents via an autoresearch loop, automating the research cycle for discovering better strategies and keeping the simulator faithful to production behavior. Safeguards against specification gaming and performance improvements to the simulator itself are key components of this evolution.

All Grab trails
  1. Powering AI-led research through simulationThis post introduces sim-rs, a new simulator for marketplace operations that runs in a single compiled program. It models bookings, drivers, trips, and time ticks, with pluggable stages for demand arrival, supply movement, cancellations, dispatch (batching, allocation, recycling), and outcome recording. The simulator supports reproducible, self-contained runs and includes an integrated experimentation platform for A/B/n comparisons. It is designed to be AI-friendly, with agents driving research loops to discover better strategies and maintain simulator fidelity. Safeguards like immutable core data, computed verdicts, scoped changes, and dual-level checks are implemented to prevent specification gaming. Performance improvements to the simulator's eligibility checks and cost calculation paths are also detailed.