Skip to main content
Back
55 XP

Lesson 8 · 16 min

Capstone: a type-safe LLM API in production

Build a complete contract-intelligence endpoint: schema, extraction with retries, streaming, and a golden-dataset accuracy gate.

What we're building

A contract-intelligence endpoint that extracts structured data from legal contracts. This is the canonical structured-output use case: deterministic schema, real downstream consequences if wrong, high value per extraction.

Requirements:

  • Accept raw contract text via POST
  • Return structured JSON: parties, effective date, termination clauses, governing law
  • Validate every field — reject partial output
  • Retry up to 3 times on validation failure
  • Pass a golden-dataset accuracy test at ≥ 85%