ConvoConnect: an app enhancing bonds through LLM-gamified conversations

Abstract

In today’s digital era, opportunities for face-to-face interactions are often replace by digital communication. Conversation card games are a good way to bridge the gap, but they are limited by static prompts, lack of personalization and the inability to adapt to players’ context. This project introduces ConvoConnect - a responsive, web-based gamified conversation application that leverages Large Language Models (LLMs) to generate personalized, context-aware conversation prompts for group or pair bonding sessions. The application offers two main modes: Casual Mode, which provides light-hearted conversation prompts and Discussion Mode, which supports more structured conversations using a memory-enabled agentic LLM architecture. The frontend of ConvoConnect is built using the ReactJS framework with TypeScript to provide a type-safe and responsive user interface. ConvoConnect adopts a microservice backend architecture, comprising three services: Casual Mode LLM, Discussion Mode LLM and User Database Service. The LLM services are developed using FastAPI, LangChain and LangGraph in Python, while the User Database Service is built using NodeJS and ExpressJS connected to a MongoDB database. Deployment platforms used for ConvoConnect include Vercel, Render and Microsoft Azure. Testing and evaluation indicate that ConvoConnect is well-received by users and effective in enhancing engagement, reducing conversational awkwardness, and facilitating dynamic, meaningful discussions. Future iterations could explore social features, technical optimizations, and broader applications in educational and professional settings.