The Challenge: A System Drowning in Documents

HM Courts and Tribunals Service (HMCTS) processes over 8 million paper forms and millions of hours of audio hearings annually. For judges, legal advisors, and case workers, the "administrative tax" of justice was reaching a breaking point.

The core challenge wasn't just digitisation—it was distillation. Stakeholders were spending 40% of their time manually transcribing hearings, cross-referencing unstructured evidence, and answering repetitive legal procedural questions. HMCTS needed a way to move from simply storing data to understanding it instantly.

The Vision: AI as a Judicial Co-Pilot

ScrumConnect partnered with HMCTS to move beyond legacy stability and into the era of Intelligent Justice. Our goal was to leverage Generative AI and Large Language Models (LLMs) to create a "digital assistant" capable of:

1. Summarising complex, multi-hour hearing transcripts into actionable case notes.
2. Surfacing specific legal precedents and procedural answers from 300+ unstructured policy documents.‍
3. Reducing the cognitive load on judges, allowing them to focus on the "human" element of the law rather than the paperwork.

‍

The ScrumConnect Approach: One Team, Responsible Innovation

1. Summarisation that Understands Law

Traditional AI often misses the nuances of legal "terms of art." We implemented Transformer-based abstractive summarisation models (using Azure OpenAI). Unlike basic tools that just "cut and paste" sentences, our solution understands the context of a witness statement or a judicial ruling, producing concise summaries that retain 100% of the legal reasoning.

‍

2. The Legal Knowledge Assistant

We built a Generative AI knowledge retrieval tool. Instead of staff scrolling through hundreds of PDFs to find a specific tribunal rule, they can now ask: "What are the current guidelines for child arrangement disputes in X chamber?" The AI interrogates the internal knowledge base and returns a cited answer in seconds.

‍

3. Guardrails First

In the justice system, "hallucinations" aren't just errors—they are risks to the rule of law. We built a Human-in-the-Loop (HITL) framework. AI-generated summaries are flagged for judicial review, and every answer provided by the knowledge assistant includes a direct "deep link" to the source document for verification.

‍

The Results: Justice at the Speed of Now

• 40% Reduction in Review Time: Judges and legal clerks reported saving nearly half their document review time through AI-powered summaries.
• Instant Knowledge Retrieval: What used to take 20 minutes of manual searching now takes under 10 seconds via the natural language assistant.
• 100% Auditability: Every AI interaction is traced and cited, ensuring the transparency required for public trust.
• From Admin to Advocacy: By automating the "typing and tidying," we unblocked frontline staff to focus on reducing the case backlog.

‍

"Our work wasn't about replacing the judge; it was about replacing the judge's highlighter. By using AI to handle the 'heavy lifting' of data processing, we’ve empowered the human experts at HMCTS to do what they do best: deliver fair and timely justice."— Laxman Kerai, Scrumconnect HMCTS Lead

What’s Next?

Building on the success of these pilots, we are looking to scale AI-enabled search across the Common Platform, creating a unified, "searchable" memory for the entire UK criminal justice system.

‍

Key Tech Stack:

• Azure OpenAI (GPT-4o) for abstractive summarisation.‍
• RAG (Retrieval-Augmented Generation) for internal knowledge management.‍
• Python & Azure Cognitive Services for speech-to-text transcription.‍
• User-Centred Design (UCD) to ensure the interface met the specific needs of the Judiciary.

‍