Capí

Capí fights environmental and climate disinformation through an AI-powered chatbot. Created by Ambiental Media and developed by journalism experts, climate science, and AI, the tool provides reliable and up-to-date information on climate change, its impacts, and mitigation actions.

How It Works

With a robust database curated by expert journalists and advanced language models like Gemini, Capí uses accessible language to democratize scientific knowledge about the climate, focusing especially on journalists, educators, and young students. In addition to being a reliable source of information, Capí serves as an interactive educational tool, promoting dynamic learning about environmental issues and contributing to a more sustainable future.

Capí’s main features include access to accurate information powered by a robust database and advanced language models, constant journalistic curation by a team specialized in environmental issues, public engagement through simple and direct language, and its use as an educational tool in schools to provide interactive learning about environmental and climate change issues. These features ensure that Capí offers reliable, up-to-date, and accessible content, making it a valuable source of information and an effective pedagogical resource for environmental awareness and education.

Documentation

Capí is a cloud-based virtual assistant developed on the Google Cloud Platform (GCP), using advanced artificial intelligence to provide accurate answers on environmental topics. Its architecture consists of three main layers: the presentation layer, responsible for the interface and processing user requests; the AI layer, which integrates the Gemini language model and the RAG (Retrieval Augmented Generation) semantic search engine to generate personalized responses; and the infrastructure layer, which uses managed GCP services such as Cloud Run and LowCO2 servers, ensuring scalability, energy efficiency, and sustainability. The RAG technique combines text generation with the retrieval of relevant information from databases, optimizing the accuracy and relevance of responses.

Additionally, the system adopts prompt tuning to efficiently adjust the behavior of the language model, reducing computational costs and energy consumption by avoiding complete retraining. This approach facilitates the customization of the agent for different contexts and continuously improves its performance. With a focus on sustainability, Capí uses servers in regions with low carbon footprints and implements optimized solutions to balance technological efficiency with reduced environmental impact. This combination makes Capí an innovative and sustainable tool for disseminating knowledge about environmental issues.

To learn more, visit the Github repository and explore the project history and system architecture.