Logo image
Back
Multi-Agent System with Generative AI and Deep Reinforcement Learning for Adaptive Financial Education
Conference paper   Open access   Peer reviewed

Multi-Agent System with Generative AI and Deep Reinforcement Learning for Adaptive Financial Education

Paulo Cesar Oliveira Brito, Joaquim Honorio, José Antão Beltrão Moura, Christian M Jones and Uwe Terton
Proceedings of the 18th International Conference on Computer Supported Education - Volume 2: CSEDU, pp.1313-1324
International Conference on Computer Supported Education, 18th (Benidorm, Spain, 18-May-2026–20-May-2026)
Scitepress Digital Library
2026
DOI: https://doi.org/10.5220/0015042800004021
pdf
CSEDU paper 2026 150428450.02 kBDownloadView
Published Version Open Access CC BY-NC-ND V4.0

Abstract

Financial Education Multi-Agent Systems Deep Reinforcement Learning Generative AI Large Language Models Conversational Agents
This paper presents a multi-agent conversational system for adaptive financial education that integrates Gen-erative Artificial Intelligence and Deep Reinforcement Learning (DRL). The architecture combines large language models (LLMs) for intent detection and sentiment analysis with a DRL agent designed to optimize educational mission sequences based on each user's profile, engagement, and context via WhatsApp. The main contribution is the design and formalization of this integrated architecture—encompassing multi-agent orchestration , MDP-based adaptive sequencing, and deployment on a widely used messaging platform—supported by a modular, incremental implementation strategy for transcultural adaptation across Lusophone countries. We present the complete system design and report preliminary empirical results from the first implementation phase (deterministic conversational flow) with Brazilian adults, which demonstrate the system's viability and effectiveness, yielding statistically significant improvements in financial knowledge, engagement, and user satisfaction compared to traditional educational materials. These findings establish a validated baseline for the subsequent phases incorporating adaptive personalization via DRL.

Details

Metrics

1 Record Views
Logo image