Morad.IA

Morad.IA: Democratizing Home Ownership through AI

Morad.IA is a comprehensive multi-agent virtual assistant designed to democratize homeownership for low-income and informal workers in Brazil by simplifying the “Minha Casa Minha Vida” (MCMV) program bureaucracy through a familiar WhatsApp interface. The solution automates the assembly of a “Financing Dossier” by extracting user data from natural language and document photos, validating eligibility against current program rules, and providing personalized real estate options.

Creativity, Social Impact, and Technical Feasibility

• Creativity: Morad.IA transforms dense, intimidating legal manuals into a warm, empathetic dialogue that uses “active offers” to show users that their dream of owning a home is mathematically possible.
• Social Impact: By targeting the “information deficit” among families in MCMV Faixas 1, 2, and 3, the project provides free technical assistance to those who typically lack access to specialized real estate consultancy.
• Technical Feasibility: The architecture utilizes a production-grade stack, including LangGraph for conversational state management, FastAPI for high-performance WhatsApp integration, and Gemini 2.5/3.1 models for high-speed document processing and reasoning.

Problem Solving and Team Collaboration

The project solves the exclusion caused by the complexity of housing rules, which often prevents eligible families from accessing government subsidies due to a lack of clear guidance on documentation and income composition. To build this prototype, the team collaborated across four specialized roles:

• The Integrator: Established the infrastructure using ngrok and Twilio to connect the WhatsApp interface to the backend.
• The Architect: Designed the LangGraph workflow to manage nodes for legal research and user interaction.
• The Mathematician: Developed the OCR vision tools and exact financial calculation scripts for credit simulation.
• The Lawyer: Curated the Retrieval-Augmented Generation (RAG) database by processing official government PDFs into searchable vector embeddings.

Technical Implementation

Morad.IA leverages Gemini 2.5 Flash for high-speed OCR to extract data from documents like RGs and pay stubs. It utilizes ChromaDB as a vector store to ground the agent’s responses in official manuals, ensuring technical accuracy through a RAG pipeline. The Streamlit dashboard provides real-time monitoring of the user’s dossier progress for brokers.

The Recommendation System

The system features a robust recommendation tool, buscar_imoveis_compativeis, which calculates a user’s purchasing power based on the rule that monthly installments should not exceed 30% of their gross income. It then cross-references this estimated “buying power” with a localized database of properties—such as the Residencial Silverstone or Recanto do Limoeiro—to suggest homes that specifically fit the user’s financial profile and geographic region.

While the team brainstormed and finalized the core concept a week prior to the event, and gathered the official Minha Casa Minha Vida manuals the day before to ensure data readiness, absolutely no code was developed before the hackathon began. The entire functional prototype—including the LangGraph orchestration, Gemini-powered OCR, and FastAPI integration—was built from scratch during the competition to transform those initial discussions into a working solution.