Generative Agents
A multi-agent simulation of customer and shopkeeper conversations, built to study how tone, politeness and empathy change how a customer feels.
Does the way a shopkeeper talks change how a customer feels?
I built AI agents to act it out and measured the customer’s sentiment. Each agent carries its own context, so conversations stay dynamic and in character, and data from the runs fed back into the agents to make them more realistic.
- Customer and shopkeeper agents
- Individual context for each agent
- Context-aware conversations via the OpenAI API
- A Pygame world for the simulation
- Data-driven tuning for realism
- Type
- Research project
- Agents
- Customer, shopkeeper
- Stack
- Python, OpenAI API, Pygame
- Studies
- Tone, politeness, empathy
Two agents, one counter
Agents with their own context
Customer and shopkeeper agents built on the OpenAI API, each with its own context.
Live conversations
Conversations play out turn by turn, reacting to what the other agent just said.
The variables
Tone, politeness and empathy on the shopkeeper’s side, measured against the customer’s sentiment.
Refined by data
Analysis of the runs fed back into the agents to make them more realistic.