Document Processing
EvenUp needed to turn huge, inconsistent personal-injury case files into accurate demands, summaries, and discovery documents fast enough for smaller firms to compete, without letting polished but incorrect reasoning slip through.
Reported impact
EvenUp reduced drafting from eight to fifteen hours of skilled legal work per case to finished drafts in about 30 minutes, raised settlement offers by 300 percent, helped customer firms clear backlogs and triple demand output, and improved smaller-case expense accuracy and overall draft completeness.
EvenUp built AI Drafts on the Claude Platform as a multi-stage document pipeline for personal-injury work. Case files often exceed 1,000 pages and include medical records, billing ledgers, police reports, scanned bills, and checkbox forms. The system climbs a cost ladder: it reads native text first, uses standard OCR where possible, and escalates to Claude vision only for the pages other methods cannot handle. Claude Sonnet and Claude Haiku manage routine extraction, while Claude Opus 4.8 handles the deeper reasoning needed to keep a thirty-page demand internally consistent and legally persuasive. EvenUp layers PI-specific data, routing, verification, and automated workflows on top, so facts get normalized once and then reused across demands, negotiation sheets, and discovery responses. Internally, teams also use Claude Enterprise and Claude Code for forecasting and debugging-agent projects.
Medical records, billing ledgers, police reports, and scanned forms enter AI Drafts; the pipeline reads native text first, then OCR, then Claude vision when needed, and routes the extracted facts into Claude reasoning layers that assemble demands, negotiation sheets, and discovery documents with reusable structured evidence.
A case often arrives as 1,000 pages or more
Documents drafted in minutes, down from 8 to 15 hours of skilled work per case
Drafting that once took 8 to 15 hours of skilled work now returns as finished drafts in about 30 minutes
Everyday extraction runs on Sonnet or Haiku. Opus takes the jobs where polish could hide an error
Raised settlement offers 300% for firms presenting the full damages picture with cited evidence
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Anthropic Sample Archived · Aug 29, 2026 Spellbook Spellbook needed to review, redline, and route very large volumes of contracts fast enough for lawyers and legal teams, while keeping issue spotting consistent and aligning recommendations with each team's preferred… 10x faster agreement turnaround time Claude Code · Claude Platform · Fable 5 
Google Cloud Sample Modified · Oct 9, 2026 Anthropic During large-scale training runs, traffic is highly variable—dropping to near zero during model checkpointing before bursting to millions of events per minute. Anthropic needed a serverless messaging queue that scaled… Handles peak throughput of over 1M events per second BigQuery · Bigtable · Cloud Pub/Sub 
NVIDIA Sample Sep 22, 2026 Sluicebox Sluicebox needed to turn fragmented bills of materials, supplier records, PDFs, CAD files, and emissions inventories into timely product carbon assessments, but manual life-cycle-analysis work was slow, expensive, and… 51-80% lower inference costs Lucy · NVIDIA Nemotron 3 Ultra · carbon graph 