CasePilot Legal AI
Plaintiff-side case intelligence with citation-grounded work product
Legal document intelligence prototype for personal-injury matters that builds a cited medical chronology, spots treatment gaps and contradictions, scores evidence coverage, models damages, and drafts demand and deposition material.
Visual evidence
Litigation review workspace
Supplementary evidence
Timeline and case risks
Supplementary evidence
Second sample matter: premises liability
Case study
Problem, Build, Outcome
Problem
Plaintiff firms spend hours reading records to build a chronology and still miss gaps or contradictions that the defense will use.
Build
Built a Node.js engine over JSON case bundles that extracts the chronology, flags issues, scores liability, causation, damages and future-care coverage, models specials and future care, and answers questions with source citations.
Outcome
Every timeline entry, risk and drafted paragraph points back to a document ID, and a single workspace turns the file into a strategy memo, damages model and downloadable case report.
Architecture
System Flow
Architecture Notes
- Deterministic retrieval and extraction keep every output reproducible and testable.
- The engine is separate from the UI so a hosted LLM or embedding layer can be swapped in later.
- A sticky section nav with scroll-spy keeps the ten-panel workspace navigable.
Engineering choices
Technical Decisions
Made citations mandatory on every generated claim rather than optional.
Kept the stack dependency-light (Node HTTP server, vanilla JS) so reviewers can run it in seconds.
Modeled future care separately from incurred specials so exposure is explainable.
Evidence
Proof Points
Treatment-gap, contradiction and missing-support detection
Evidence coverage matrix with strength scoring
Citation-grounded demand summary, deposition outline and Q&A
Next read