How rohasnagpal/vertexpay-indian-legal-ai-benchmark works
VertexPay is a document-based legal AI benchmark, not an executable application. Its principal workflow is to give an AI the participant case materials, ask it to produce a cited chronology and analysis, then evaluate the answer against benchmark tasks and scoring guidance. The graph groups the case evidence by legal function and keeps instructor-only answers separate. Relationships are based on the README; no source excerpts were supplied, so internal document-level flows are not inferred.
Main entry points: participant Participant case file · SCORING.md Scoring guide · instructor_benchmark_guide.pdf Answer key · TASKS.md Benchmark tasks
Components 13
Participant evidence
- Contracts and terms
participant/01_shareholders_agreement.pdf - Corporate governance
participant/05_board_papers_and_minutes.pdf - Messages and emails
participant/07_founder_email_chain.pdf - Financial records
participant/17_cap_table_and_dilution.xlsx - Legal proceedings
participant/13_nclt_company_petition.pdf - Participant case file
participant
Benchmark workflow
- Benchmark tasks
benchmark/TASKS.md - Cited analysis
Instructor evaluation
- Scoring guide
benchmark/SCORING.md - Answer key
instructor/instructor_benchmark_guide.pdf - Benchmark results
Other components
- Benchmark user
- AI system
Connections 8
- Benchmark user → Participant case file: selects materials
README.md - Participant case file → AI system: supplies evidence
README.md - Benchmark tasks → AI system: sets task
README.md - AI system → Cited analysis: produces analysis
README.md - Cited analysis → Scoring guide: evaluated with
README.md - Cited analysis → Answer key: compared against
README.md - Scoring guide → Benchmark results: scores answer
README.md - Answer key → Benchmark results: informs scoring
README.md