Fogo Cruzado, a Brazilian organisation that monitors armed violence, has built one of the most extensive datasets on shootings in Latin America. Since 2016, it has recorded more than 63,000 incidents across Rio de Janeiro, Recife, Salvador and Belém. Yet data alone is not enough to create impact. To address the challenge of accessibility, the organisation developed VANIA, a chatbot powered by artificial intelligence that allows citizens to ask simple questions and get real-time information on gun violence in their communities.

The problem: Shedding light on the violence  

Gun violence is a daily reality in Brazil’s major cities, particularly in poor and marginalised neighbourhoods. In 2024 alone, 2,922 people were killed by police forces in the metropolitan regions covered by Fogo Cruzado. But the lack of accessible data weakens public debate and prevents communities from demanding accountability.

“We have the biggest database on armed violence in Latin America, but despite its public importance, we still face the major challenge of making it accessible,” explains Diogo Santos, communication and innovation manager at Fogo Cruzado.

Building the solution: Roadmap to prototyping

The initial goal was simple: allow citizens to query Fogo Cruzado’s database through a conversational interface. Early prototypes were tested in Portuguese and English, offering suggestions of possible questions to guide users. The tool enables follow-up queries on the same topic, simulating a natural conversation.

What seemed straightforward quickly revealed technical complexities. The team had to adapt the LLM to Fogo Cruzado’s unique taxonomy, including terms like “police massacre” (three or more civilians killed in an operation) or “drive-by attack”. Training the system to understand this specialised vocabulary was a critical step.

The team, technologies and the skills behind the project

VANIA combines a LLM with Fogo Cruzado’s API, which feeds real-time information from the database. The chatbot checks each query against the database and validates the result before responding. The interface was designed for simplicity, showing users what kind of questions can be asked and how to refine them. A bilingual version is currently being tested, with Portuguese as the primary language.

The project was led by Fogo Cruzado’s communication and innovation team, working closely with developers and AI specialists. Their expertise in database management and violence monitoring was crucial to adapting the model to their dataset. The team also relied on community knowledge to refine terminology and ensure that the language used by the chatbot matched the reality of affected populations.

Developing VANIA highlighted several challenges:

• Specialised vocabulary: Training the LLM to recognise and explain local terms and categories required significant effort.

• Cost barriers: As Santos notes, “How do we democratise the use of AI and apply it to real problems when cost is a big barrier?” Tools priced in US dollars remain inaccessible to many Latin American organisations.

• Naming and cultural fit: The project was originally called BulletData, but the team quickly realised it would not resonate locally. They opted for VANIA, short for Violência Armada em Números + IA (Arms violence in numbers + AI), a name that is both meaningful and familiar to Brazilians.

The opportunities: Transparent and evidence-based conversations on violence

For Fogo Cruzado, VANIA is more than a chatbot: it is an entry point to a future ecosystem of tools for transparency and advocacy. By making gun violence data accessible, it empowers citizens to stay informed, organise collectively, and pressure authorities to act. The tool also has potential for journalists and policymakers, who can use it to identify trends and design evidence-based responses. Beyond Brazil, the model could inspire other contexts where violence is poorly documented and misinformation thrives.

Lessons for newsrooms

• Accessibility matters as much as data: Collecting information is not enough. To empower communities, data must be translated into usable, everyday formats.

• Local language is key: Training AI systems on context-specific terminology ensures accuracy and cultural relevance.

• AI can support civic mobilisation: Tools like VANIA show how AI can be harnessed not only for efficiency, but to strengthen democracy and social accountability.