DEV Community

CourtGPT
CourtGPT

Posted on

Improving Pro Se Litigant Access Through AI: A Working Pattern

#legaltech #ai #access #research

Improving Pro Se Litigant Access Through AI: A Working Pattern

The Legal Services Corporation's 2022 Justice Gap report finds roughly 75% of U.S. civil legal needs go unmet. This article summarizes working deployment patterns that have improved legal access for self-represented litigants.

Why AI can help

Pro se litigants typically need:

  1. Procedural guidance (forms, deadlines)
  2. Plain-language rights explanations
  3. Document drafting help
  4. Lawyer referral when appropriate

AI can address all four without replacing attorneys.

Working patterns

Public self-help portal

Source deployment pattern:

  • Self-hosted RAG over jurisdiction corpus
  • BAA-protected LLM endpoint
  • Plain-language response + citation to source
  • Disclaimer every page: "This is not legal advice"
  • Lawyer escalation for time-sensitive issues
  • Mobile-first UI
  • Multilingual support

Internal research tool (for legal aid lawyers)

  • SSO via Okta / Entra
  • Self-hosted OpenSearch + pgvector
  • BAA endpoint (Azure OpenAI / AWS Bedrock)
  • Audit logging
  • Citation verification

Court clerk assistant

  • On-prem deployment
  • Fine-tuned Llama 3.3 70B on court rules
  • Strict access controls
  • Disclosure to bench

Real-deployment examples (not endorsements)

  • Texas Law Help (texaslawhelp.org) - state bar portal
  • LawHelp.org - LSC-supported
  • Upsolve (upsolve.org) - Chapter 7 specific
  • MassLegalHelp (masslegalhelp.org)
  • Stanford RegLab/HAI California Supreme Court collaboration

Failure modes observed

  1. Tone too formal (cited statute, no action)
  2. No source documents
  3. No path to a lawyer
  4. No mobile UI
  5. No language access

Eval framework

Track monthly:

  • Helpfulness rating (thumbs up/down)
  • Resolution rate (% who completed form / found answer)
  • Escalation rate (% who sought a lawyer afterward)
  • Citation verification rate
  • Hallucination audit (100 random responses reviewed by lawyer)
  • Equity audit (response quality by language + demographic)

Open-source references

Limits of AI deployment

Critical caveats:

  1. Cannot replace attorney judgment
  2. Cannot serve emergency needs (DV, eviction, child custody)
  3. Cannot handle complex document review

AI works best for T1 questions: "What form do I file?", "What's the deadline?", "What's the statute?" (60-70% of pro se traffic in deployments).

Acknowledgments

This article summarizes public research and working patterns. Specific deployments vary by jurisdiction.

Dillon Deutsch has built legal research AI systems serving users in 50 US states. https://courtgpt.ai

Top comments (0)