The Mall Recommendation Multi-Agent System is an AI-driven framework built on a multi-agent architecture to enhance retail experiences in shopping malls. It consists of shopper agents that track visitor interactions, preference agents that analyze past and real-time data, and recommendation agents that generate tailored product and promotion suggestions. Agents communicate via a message-passing protocol to update user models, coordinate cross-agent insights, and adjust recommendations dynamically. The system supports integration with CMS and POS for real-time inventory and sales feedback. Its modular design allows developers to customize agent behaviors, integrate new data sources, and deploy on various platforms. Ideal for large retail environments, it improves customer satisfaction and boosts sales through precise, context-aware recommendations.

Duet GPT is a Python-based open source framework for orchestrating multi-agent conversations between two GPT models. You define distinct agent roles, customized with system prompts, and the framework manages turn-taking, message passing, and conversation history automatically. This cooperative structure accelerates complex task resolution, enabling comparative reasoning, critique cycles, and iterative refinement through back-and-forth exchanges. Its seamless OpenAI API integration, simple configuration, and built-in logging make it ideal for research, prototyping, and production workflows in coding assistance, decision support, and creative ideation. Developers can extend the core classes to integrate new LLM services, adjust the iterator logic, and export transcripts in JSON or Markdown formats for post-analysis.

The Java Action Linearprogram module provides a specialized action for the LightJason framework that allows agents to model and solve linear optimization tasks. Users can configure objective coefficients, add equality and inequality constraints, select solution methods, and run the solver within an agent’s reasoning cycle. Once executed, the action returns the optimal variable values and objective score which agents can use for subsequent planning or execution. This plug-and-play component abstracts solver complexity while maintaining full control over problem definitions through Java interfaces.