Simulation
Test your agents with simulation
Overview
For conversational agents, a scenario is often a dynamic sequence of interactions between the agent an the user.
Simulation allows you to go beyond static test cases and verify your conversational agent against a dynamic testbot that simulates a user.
Example: Customer Support Simulation
In the following example, the pink bot is the customer support agent in test and the yellow bot is the simulated user testbot.
The testbot is given the task (e.g. refund, produt question), set of information (such as account, order number, etc.), and personality (e.g. friendly, anxious, etc.).
The testbot is an LLM-based conversational agent itself and behaves non-deterministically. To qualify if a simulation run is a valid test run (if the testbot behaved correctly), we can use the precondition
in the contract and only consider whether the agent-in-test satisfies the contract in runs where the preconditions are met.
Want to use simulation for your agent?
Relari provides simulated testbot templates for a variety of conversational agent use cases. Please contact us at founders@relari.ai to get access.