Legal Help Commons / Working Groups

Build Together

The Legal Help Commons organizes working groups around the shared technical challenges that every legal aid organization faces when adopting AI. Each group develops practical playbooks, reference architectures, and tested tools — so nobody has to solve these problems alone.

🎙

Voice AI Intake & Triage

Reach people where they are

Not everyone interacts with legal help through a screen. Many of the people who most need help — older adults, people with limited English proficiency, those in crisis, people without reliable internet — are more likely to pick up a phone than visit a website. If AI-powered legal help only works through chat and web interfaces, it misses the people who need it most.

This group explores how voice-based interfaces — phone systems, voice assistants, and conversational AI — can serve as an intake and triage channel for legal help. We're tackling the hard design questions: how do you gather enough information through voice to route someone correctly? How do you handle the messiness of spoken language, accents, and emotional distress? How do you make it accessible and safe?

What This Group Covers

  • Speech-to-structured-data pipelines for legal intake
  • Voice interaction design for distressed and diverse callers
  • Triage logic for phone-based channels
  • Multilingual and accessibility considerations
  • Integration with existing hotlines and IVR systems

What This Group Produces

  • Design patterns for voice-based legal intake
  • Reference architecture for voice-to-triage systems
  • Evaluation framework for voice AI accuracy and safety
  • Implementation playbook for legal aid call centers
  • Tested prompt strategies for voice-based issue spotting
📄

Data Extraction from Documents & Images

Read the documents

Legal help workflows constantly encounter documents — court filings, summonses, demand letters, handwritten forms, faxes, blurry phone photos. Before any AI tool can help a person understand or respond to a legal document, it first has to reliably extract structured data from it. This is harder than it sounds: legal documents come in wildly different formats, contain critical details where small errors have outsized consequences, and often include personally identifiable information that must be handled with care.

This group brings together technologists from Stanford, Ohio Legal Help, Pew/Massachusetts Courts, LSC, Duke, and others who are all solving variations of the same extraction pipeline problem. We've already produced a comprehensive OCR playbook, a reference architecture, and a PII masking protocol — and we continue to refine and extend these tools.

What This Group Covers

  • OCR tool evaluation and selection (Azure, Tesseract, Llama Vision, etc.)
  • Schema-first extraction with LLM-powered field mapping
  • Validation loops and error handling for legal-critical data
  • PII detection, masking, and safe processing strategies
  • Handwritten document and degraded image handling

What This Group Produces

  • OCR & Data Extraction Playbook (published, continuously updated)
  • Reference architecture with Mermaid diagrams and code patterns
  • PII Masking Protocol for legal document processing
  • Starter prompt library for structured extraction
  • Evaluation rubric and test cases for extraction accuracy
📚

AI-Ready Knowledge Bases

Build the foundation

The most sophisticated AI pipeline in the world is useless without authoritative, jurisdiction-specific content to ground it. When someone asks a legal help chatbot about eviction rights in Ohio, the answer needs to come from Ohio-specific legal content — not a hallucinated generalization. Building knowledge bases that are structured, labeled, current, and AI-retrievable is the essential foundation for every AI tool in legal help.

This group works on the full knowledge base lifecycle: structuring legal help content with consistent metadata, exporting it from CMS platforms (Drupal, WordPress, custom systems), indexing it for retrieval-augmented generation, keeping it current as laws change, and evaluating whether AI tools actually use it correctly. Members span legal aid organizations, courts, and technology partners across multiple states.

What This Group Covers

  • Content structuring standards and metadata schemas
  • CMS export strategies (Drupal, WordPress, and others)
  • RAG architecture patterns and embedding strategies
  • Multi-jurisdiction content management
  • Content freshness, versioning, and maintenance workflows

What This Group Produces

  • Knowledge Base Building & Use Playbook
  • Content structuring standards and metadata schema
  • Reference RAG architecture (including Ohio's Drupal integration)
  • Content readiness checklist for legal aid organizations
  • Evaluation rubrics for KB-grounded AI accuracy

Who Should Join?

These groups are open to anyone working on AI adoption in the access to justice sector. You don't need to be a technical expert — you need to be working on the problem.

Legal Aid Technologists

Building or maintaining AI-powered tools for your organization. Contribute your real-world implementation experience.

Content & Knowledge Managers

Managing legal help websites, guides, and directories. Your content expertise is the foundation AI depends on.

Court IT & Innovation Teams

Working on document processing, online services, or data systems at courts. Share challenges and solutions with peers.

Researchers & Academics

Studying legal AI, NLP, or access to justice. Bring methodological rigor and help build evaluation frameworks.

Technology Partners

Building tools for the legal aid market. Align your products with community standards and get direct user feedback.

Program Staff & Practitioners

On the front lines of legal help delivery. Your perspective on what users actually need keeps the work grounded.

Join a Working Group

Sign up to receive information about upcoming meetings, shared resources, and how to contribute. You can join one group or all three.

Sign Up via This Form →