Skip to content

Justice · State public sector

AI analysis of court hearings: from courtroom video to the minute in minutes

3 h → 40 min
from listening to a full hearing to having its analysis ready to consult
5
areas of law in a single platform: civil, criminal, family, commercial, and labor
8
AI steps per hearing, with no human intervention after registration
209
speaker-attributed interventions in a 43-minute hearing

The challenge

A state justice institution holds thousands of hearings across five areas of law —civil, criminal, family, commercial, and labor—. Every session was recorded, but reviewing a three-hour hearing meant listening to all of it; transcripts were done by hand or not at all; each court worked in isolation, and there was no way to measure times, caseload, or traceability of what happened in the courtroom.

Centralize the documentary and audiovisual record of the courts and turn, fully automatically, every recording into structured, actionable information —a speaker-attributed transcript, a minimum record, procedural actors, alerts, and metrics— without adding any workload to court staff beyond registering the hearing.

The solution

A responsive web platform built 100% on Google Cloud, made of decoupled, specialized services organized in layers: judicial management, ingestion and video, and artificial intelligence. When a session ends, the video is ingested automatically into cloud storage; a speaker-identification transcription engine (diarization, in two profiles depending on length) turns the audio into text attributed to each actor; and an orchestrator triggers a Gemini AI pipeline that, per hearing, generates the minimum record, identifies the actors, analyzes sentiment and conflict, raises alerts, computes dashboard metrics, and vectorizes the content for semantic search. A conversational legal assistant (RAG) answers in natural language about case files, hearings, codes, and case law. All under role-based access control (RBAC), jurisdiction-level data segregation, full action auditing, and credentials in a secrets manager.

  • Google Cloud
  • Gemini
  • Speaker-diarized transcription
  • Semantic search (RAG)
  • Vector database
  • Microservices
“Staff only register the hearing; by the time they come back to the system, the transcript, the record, and the alerts are ready to review.”
— Innovation team · state justice institution

Case described by sector for confidentiality. Figures are operational indicators of the implemented system; the data in the associated lab is 100% synthetic.

Service used

Generative AI & agents

Gemini, Vertex AI and RAG solutions that answer with verifiable sources.

Is your organization facing something similar?

Tell us your challenge and we'll propose a concrete path — no strings attached.

Contact Nuuptech Try it yourself: live instruments in the lab →