LegisLight

AI for Better Governance

Comment

Team Members: Izabela Litwinowicz, Michelle Shi, Remi Ladia

Category Tags: Community Resilience and Accessibility

Github Link: https://github.com/rladia/Claude-Builders-Club-Hackathon/tree/main

Video Link: https://youtu.be/9JRNNWPWM84

What Problem Are We Solving?

The legal system is built on principles of fairness and equal access, yet in practice:

1. Legal language is extremely complex and inaccessible

  • Statutes, regulations, court filings, and legal notices are written at a reading level far above that of the general public.
  • Most people—including many who must navigate the legal system—cannot understand:
    • what they are being charged with
    • what rights they have
    • what obligations or deadlines apply to them
    • what a court document actually means
  • This creates confusion, fear, and unintentional non-compliance.

2. People with limited English proficiency face even greater barriers

  • In many jurisdictions, 40–60% of defendants speak English as a second language or not at all.
  • Court notices and legal paperwork are rarely:
    • translated
    • simplified
    • culturally adapted
    • explained clearly
  • This results in miscommunication, missed deadlines, and violations of due process.

3. Courts are overwhelmed with high caseloads

  • Judges, clerks, and public defenders often work under intense workloads.
  • They do not have the time or resources to:
    • explain every statute
    • summarize every filing
    • read through hundreds of pages of legislation
    • give language-access support to every user
  • The consequence is a less efficient and less equitable legal system.

4. AI tools exist, but none are designed responsibly for courtroom use

  • General-purpose chatbots:
    • hallucinate legal facts
    • give unauthorized legal advice
    • summarize inaccurately
    • cannot be used in official proceedings
    • lack auditability and safety controls
  • Courts need a tool that is accurate, neutral, transparent, and court-safe.

How Our Solution Works

LegisLight is a specialized AI system designed for courtroom use, focused on transparency, fairness, and accessibility. It does not give legal advice. Instead, it provides readable, accurate, and multilingual explanations of legal texts.

1. Court Documents/Audio In → Clear Explanations Out

The system takes in:

  • bills
  • statutes
  • court notices
  • arraignment letters
  • motions
  • legal forms
  • audio recordings

Then it performs three core functions:

A. Plain-language summarization

  • Converts complex legal text and courtroom audio into clear, eighth-grade reading-level explanations.

B. Rights identification

Extracts and explains rights embedded in documents, such as:

  • right to counsel
  • right to an interpreter
  • right to remain silent
  • right to a speedy trial
  • right to understand charges

All explanations are informational, not advisory.

C. Multilingual translation

  • The output is translated into whichever language the user selects.
  • A back-translation check ensures meaning is preserved.

2. Designed with Courtroom-Grade Safety

No legal advice

  • Hard filters prevent strategic or advisory language.

Verification pass

  • A second AI verifies accuracy against the original text to prevent omissions or hallucinations.

Full audit logs

Every transformation is tracked:

  • model version
  • timestamp
  • input text hash
  • output text
  • errors and flags

Courts can reconstruct every step.

On-premise option

  • The system can run inside a courthouse network without sending data to the cloud.

3. Transparent, Explainable AI

Users always see both:

  • the original text/audio transcription, and
  • the simplified explanation

Nothing is hidden.
Nothing is fabricated.
This strengthens trust in the tool and in the judicial process.

Why It Matters

1. Expands Access to Justice

Millions of people appear in court each year without fully understanding their:

  • charges
  • notices
  • obligations
  • rights

LegisLight ensures everyone, regardless of education or language, can understand their legal situation.

2. Supports Due Process & Fairness

A person cannot exercise rights they do not understand.
By making rights clear, LegisLight strengthens:

  • procedural fairness
  • constitutional protections
  • legitimacy of court outcomes

3. Improves Court Efficiency

When defendants, litigants, and the public understand:

  • deadlines
  • requirements
  • next steps

There are fewer delays, fewer misunderstandings, and fewer unnecessary hearings.

4. Enhances Transparency of Government

  • Complex bills and regulations become understandable to ordinary people.
  • This promotes:
    • accountability
    • informed civic engagement
    • better public oversight
    • reduced mistrust in government

5. Safe, Responsible Use of AI in Sensitive Environments

Most AI tools are not appropriate for courts.
LegisLight is built specifically to be:

  • transparent
  • accurate
  • audit-ready
  • bias-mitigated
  • designed for legal neutrality

It sets a new standard for responsible legal AI.

Summary

What problem we’re solving:

The legal system is difficult to understand, especially for people with limited English proficiency or limited legal knowledge. This leads to unfair outcomes and inefficient courts.

How our solution works:

We use a legally-trained, safety-filtered AI system to summarize legal documents, identify rights, translate explanations, and operate with full auditability.

Why it matters:

This improves access to justice, strengthens due process, increases transparency, and helps courts operate more efficiently.

Citations

We used OpenAI and Claude to help create code for our project.
We also used Figma and a plugin to convert the design to React code.

Challenges we faced

  • Claude couldn’t take audio input: We had to integrate OpenAI’s speech-to-text pipeline instead and redesign our workflow to support dual-model processing.
  • Balancing accuracy with safety: Ensuring the system never gives legal advice while still providing helpful explanations required layered filters and verification.
  • Handling complex legal language: Some statutes and motions contain dense cross-references and legal definitions, making plain-language summarization difficult.
  • Multilingual consistency: Keeping translations accurate while maintaining legal meaning, especially after back-translation checks, was technically challenging.
  • UI clarity: Displaying original text, explanations, transcriptions, and logs without overwhelming users required careful interface planning and iteration.

Accomplishments we're proud of

  • Built a full, working pipeline that handles both text and audio inputs.
  • Developed a rights-identification system that stays strictly informational.
  • Designed a fully transparent, audit-ready workflow with logs, hashes, and model tracking.
  • Created a polished React frontend, converting Figma designs into functional code.
  • Successfully integrated OpenAI and Claude, each powering different parts of the pipeline based on strengths.

What We Learned

  • How to navigate the limitations of different AI providers and design a robust, hybrid workflow.
  • The importance of strict safety, transparency, and neutrality when working with legal information.
  • Best practices for FastAPI development, backend verification steps, and multi-model pipelines.
  • How to design user-friendly interfaces for people with limited legal or language proficiency.

What’s Next for LegisLight

  • Improve speaker diarization to better process multi-speaker courtroom recordings.
  • Add OCR support for scanned documents and mailed court notices.
  • Build an admin dashboard for courts to view logs, manage permissions, and monitor accuracy.
  • Enhance bias testing and mitigation through additional evaluation datasets.
  • Add a live transcription feature to summarize what is happening in court in real time, providing instant, accessible explanations of ongoing proceedings.

Built With

Share this project:

Updates