# GuaIA

> A multimodal AI system that detects electoral disinformation across text, images, video and audio in under 4 seconds. 1st place at the CNJ Innovation Prize 2025.

- Organisation: TRE-GO
- Role: Software Engineer and Co-architect
- Duration: Aug 2024 - Jul 2026
- Tools & Technologies: Python, NestJS, Multimodal AI, PostgreSQL
- URL: https://adrianoleao.com/case-study/guaia
- Author: Adriano Leão (https://adrianoleao.com/)

## Context

Electoral disinformation doesn't come in a single format: it spreads as text, images, video and audio. The Tribunal Regional Eleitoral de Goiás (TRE-GO), together with UFG, needed a system able to analyze all of them, and to do it fast, ahead of Brazil's 2026 elections.

    - **1st**: Place at the CNJ Innovation Prize 2025

    - **< 4s**: To analyze text, images, video and audio

    - **4**: Content types handled by one system

## My role

Software engineer and co-architect of the project. I shared the architectural decisions, designed and implemented the Python and NestJS services, integrated multiple databases and coordinated across the project's different technical fronts.

## What I built

- A multimodal AI system that analyzes text, images, video and audio
- Python and NestJS services working together as one product
- Integration across multiple databases
- A pipeline fast enough to return an analysis in under 4 seconds

## Outcome

- 1st place at the CNJ Innovation Prize in 2025
- Detection of electoral disinformation across four content types in under 4 seconds
- The system has since been adopted

[Back to the profile](https://adrianoleao.com/index.md)