Conversational Agent for Personalized and Interactive Recommendation of Professionals for Employers

• Grant number: 25/22069-0
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: May 01, 2026
• End date: January 31, 2027
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Eric Keiji Tokuda
• Grantee: Eric Keiji Tokuda
• Associated researchers: Anderson da Silva Soares

Abstract

This project aims to develop and validate a conversational recommendation system with artificial intelligence, to be integrated into the Famyle digital platform, which connects domestic workers to employers throughout Brazil. The proposed system will enable users to interact with an intelligent conversational agent, capable of understanding preferences, interpreting contexts, and suggesting personalized recommendations in a transparent, explainable, and accessible manner. The justification for this proposal is anchored on two pillars: (i) technological, in light of the recent evolution of Conversational Recommender Systems (CRS), with promising applications still underexplored in the context of informal e-recruitment; and (ii) social, considering that informality in domestic work in Brazil mainly affects young women, often in situations of social vulnerability and digital exclusion. The project therefore seeks to promote innovation with social impact, connecting cutting-edge technology to a traditionally neglected market. The solution will face scientific challenges such as: acquisition and curation of representative data, algorithmic explainability, diversity and fairness in recommendations, multichannel integration, and multimodal data fusion. To this end, the methodology will adopt reinforcement learning techniques, attention-based models, Maximal Marginal Relevance (MMR), and a modular architecture with RESTful APIs. The platform will use anonymized data from its active base (with more than 1.4 million users) to train and validate the models, with execution of A/B testing, engagement analysis, and usability evaluations. The team includes researchers with solid backgrounds in AI, NLP, and Software Engineering, with technical activities in MLOps and interaction design developed by the internal team, in addition to specialized scientific consultancy in algorithmic fairness and personalized recommendation. The existing computational infrastructure (cloud, APIs, database, web/mobile interface) will be sufficient for the development of Phase 1. At the end of the project, a functional system at TRL 6 is expected to be delivered, validated with real users in an operational environment, with positive metrics of accuracy, explainability, satisfaction, and retention. From a commercial perspective, the solution will have immediate application in the Famyle platform itself, as well as potential for licensing as SaaS or white label to other sectors (caregivers, operational services, productive inclusion). (AU)