What it does

lets administrators simulate scripted conversations between the AI agent (Lora) and a real member’s profile. The system processes your scripted messages as if they were that member’s inputs, records the full exchange, and extracts structured insights from the AI’s responses. No messages are ever sent to the real member.

Use AI Simulations to:

  • Verify that prompt changes produce the expected tone and content before they go live.
  • Understand how Lora would respond to a specific member’s profile and answer history.
  • Regression-test AI behavior after model or configuration updates.

Simulations are purely a backoffice testing tool — running them has no effect on the member’s app experience, questionnaire progress, or introduction history.

Simulations list

ColumnMeaning
#IDUnique simulation identifier.
ModelThe AI model used for this simulation.
ProfileThe member’s profile the simulation was run against (resolved to the member’s name).
Executed AtWhen the simulation was last run.
DescriptionAn admin-entered note describing what this simulation tests.

Filter the list by Model to compare behavior across different LLM configurations. Click any row to open the simulation detail view.

Creating a simulation

  1. Click + AI Simulation.
  2. Enter a Description for your own reference.
  3. Select the Model: one of the four available OpenAI model options. The model is locked after creation — to test a different model, create a new simulation.
  4. Enter the Profile ID of the member whose profile data the simulation should use. The profile ID is locked after creation.
  5. Click Save.

After saving, go to the Messages tab to add the scripted user-turn messages (see Running a simulation below).

Available models

Model optionNotes
openai:gpt-4.1-nanoFastest and lowest cost; suitable for high-volume or quick checks.
openai:gpt-4o-miniBalanced speed and capability.
openai:gpt-4oHigh capability; slower and more expensive than mini variants.
openai:gpt-3.5-turboLegacy model; available for comparison testing against older agent configurations.

Running a simulation

  1. Open the simulation and go to the Messages section.
  2. Add one or more messages — these are the scripted user-side turns to feed into the AI. Click Add Message, type the message text, and repeat for each turn. Drag messages to reorder them.
  3. Click Execute in the action bar. The system sends each message in order to the inference service, which processes them against the member’s real profile data using the configured AI agent prompt. The exchange is saved as a new run in the History section.

Each run adds a timestamped entry to the History section. The most recent run’s output also populates the Insights section with any structured data the AI extracted from the conversation.

Detail view sections

SectionWhat it shows
Information Description, selected model, and profile ID. Model and profile are read-only once saved.
Matches The member’s actual introductions (filtered to matched status), shown for context when testing match-related agent behavior.
Messages The ordered list of scripted user-turn messages. Add, delete, or reorder messages before running.
History Past run results, each keyed by timestamp. Shows the full message exchange and AI responses from that run.
Insights Structured data (JSON) extracted from the most recent execution — for example, personality traits, profile flags, or custom fields the AI agent is configured to identify.