Ministério Público do Estado do Rio Grande do Sul

Video evidence has become central to modern criminal prosecution, but manually reviewing it at scale creates severe bottlenecks. For prosecutors at the Ministério Público do Estado do Rio Grande do Sul — one of Brazil's largest state prosecution offices — each case could involve hours of recorded testimony, witness interviews, and surveillance footage requiring close review to identify contradictions, assess credibility, and produce accurate transcripts for the record. With no automated tooling, prosecutors spent substantial portions of their working time on rote review tasks rather than legal analysis and charging decisions. At scale, this translated directly into delayed proceedings, heavier caseloads, and reduced prosecutorial capacity across the state.

In partnership with Xertica.ai, the office deployed a generative AI platform purpose-built for legal video analysis. The solution leverages large language models to perform automatic speech transcription with speaker diarisation — distinguishing individual voices in multi-party recordings — alongside AI-generated summaries, contradiction detection across statements, and sentiment and bias analysis of video content. Xertica.ai's role covered both the model integration and deployment within the office's existing evidence workflows. The platform operates at the intersection of speech recognition and natural language processing, delivering structured analytical output from raw video files rather than requiring manual review. Deployment began in late 2024 as part of Brazil's broader prosecutorial AI expansion, which also encompasses tools like Jarvis and LuminarIA at other public ministries across the country.

The platform reduced video analysis time by up to 90%, fundamentally changing how prosecutors engage with video evidence. Between November 2024 and May 2025, the system processed over 23,400 videos, saving more than 11,500 hours of prosecutor work — equivalent to years of full-time analytical effort compressed into six months. Key outcomes include:

• 90% reduction in time required to review and analyse video evidence per case
• 23,400+ videos processed across the six-month initial deployment period
• 11,500+ prosecutor hours reclaimed for higher-value legal reasoning and charging work
• Qualitative shift: prosecutors now receive structured, searchable transcripts and contradiction flags rather than raw footage

• Generative AI can deliver step-change productivity gains in evidence review without displacing prosecutorial judgment on charging decisions — the tool surfaces insights, humans act on them.
• Speaker diarisation and contradiction detection add meaningful investigative value beyond simple transcription, justifying the deployment at high volume.
• Rapid adoption (23,400 videos in six months) demonstrates that large prosecution offices can absorb AI tooling without extended pilot phases when institutional support is in place.
• Brazil's cross-agency AI infrastructure — spanning LuminarIA, Jarvis, and now Xertica.ai — shows that platform momentum within a jurisdiction can accelerate deployment at the state level.