case-digest-2026

A 90-second AI case digest that helps personal-injury teams understand what changed, what matters, and where to look next without digging through scattered case records.

Comment

Inspiration

Personal-injury case files can become dense very quickly: notes, emails, medical records, deadlines, expenses, documents, and ongoing updates all live in different places.

We wanted to answer a simpler question:

Can someone understand where a case stands in about ninety seconds, without losing the ability to verify every fact?

Team

Built by: Michael Guzman Vanshika Srivastava Vanessa Victorino

Case Digest was built for the Swans Applied AI Hackathon at Law-Di-Gras in San Diego, October 2026.

What it does

Case Digest turns a live Clio Manage matter into a structured, source-backed dashboard for two audiences.

Firm view

The legal team can quickly see:

  • important case facts
  • deadlines
  • medical specials
  • policy limits
  • working case value
  • recent changes
  • facts ranked by importance
  • the original source behind each extracted fact

Every fact includes a Source action that opens the note, email, record, or cited document page it came from.

Provider view

Treating medical providers can see a narrower view of the case without receiving the attorney's entire file.

Only facts explicitly marked as shareable appear in this view. Sensitive information such as case value, strategy, expenses, deadlines, and coverage remains private by default.

How it works

The workflow is:

Clio Manage → read-only sync → SQLite → AI extraction → cited facts → dashboards

The Clio integration is read-only by design. Case records are pulled into a local SQLite database, where new and changed records can be processed.

Each record is hashed, so unchanged information does not need to be analyzed repeatedly.

Structured information such as tasks and expenses is handled deterministically. AI is used where semantic interpretation is more useful, such as notes, emails, medical documents, and scanned PDFs.

Scanned PDFs can be read visually, and longer documents are split into chunks while preserving page references.

Why citations matter

We did not want Case Digest to become another opaque AI summary.

Every extracted fact stores:

  • its originating record
  • a supporting quote
  • the document page when applicable

That means someone reviewing the digest can move directly from a summarized fact back to the evidence behind it.

The AI helps with comprehension, but the source remains available for verification.

Built with

  • Next.js
  • TypeScript
  • Tailwind CSS
  • SQLite
  • Anthropic Claude
  • Clio Manage API
  • Node.js
  • pdf-lib
  • mammoth

Challenges

One of the biggest challenges was deciding what should and should not be handled by AI.

Not everything benefits from a language model.

Tasks, expenses, names, dates, and other exact information can often be handled deterministically. Notes, emails, scanned documents, and narrative medical records require more interpretation.

Another challenge was building useful privacy boundaries within a short hackathon window.

The provider view demonstrates the concept of selective sharing, but the current prototype does not yet include separate provider authentication or server-side access enforcement.

We chose to document those limits rather than present the prototype as more secure or complete than it currently is.

What we learned

The project reinforced that the useful part of an AI product is not simply generating a summary.

The real product decisions are:

  • what information should be surfaced
  • what should remain private
  • what should be handled deterministically
  • when AI actually adds value
  • how users verify AI-generated information
  • how the tool fits into an existing workflow

The result became less about "AI summarization" and more about fast, traceable case understanding.

Current status

Case Digest works end to end with both a live Clio matter and a fictional seeded matter.

The prototype currently supports:

  • read-only Clio sync
  • incremental updates
  • AI-assisted extraction
  • scanned and multi-page PDF processing
  • cited facts
  • firm and provider views
  • source inspection
  • change tracking
  • cost logging

The project also includes a fictional /demo matter so the interface can be explored without exposing real case information.

What's next

A production version would need stronger access control, persistent sharing permissions, provider authentication, server-side privacy enforcement, expanded testing, and production-ready deployment architecture.

Built With

Share this project:

Updates