Scenario: Structured Data Extraction You are building a structured data extraction system using Claude. The system extracts information from unstructured documents, validates output using JSON schemas, and maintains high accuracy. It must handle edge cases gracefully and integrate with downstream systems. Your pipeline extracts obligations from vendor contracts, policy memos, and meeting notes into a consolidated compliance matrix. Reviewers find that when two sources support the same obligation with slightly different wording, final rows contain a polished obligation statement and a bibliography list, but no row-level evidence trail. Auditors reject several rows because they cannot verify which excerpts were merged into each statement. What should you change?
Context Management and Orchestration
Loading lesson content...
Scenario: Structured Data Extraction You are building a structured data extraction system using Claude. The system extracts information from unstructured documents, validates output using JSON schemas, and maintains high accuracy. It must handle edge cases gracefully and integrate with downstream systems. Your structured data extraction pipeline processes invoices from multiple vendors. Each invoice goes through three subagents in sequence: a document parser, a field extractor, and a schema validator. During a production incident, you discover that when the field extractor encounters a timeout connecting to an internal enrichment service, it catches the exception and returns an empty extraction result marked as successful. The schema validator then flags this empty result as a validation failure. The coordinator receives a generic "schema validation failed" error and retries the entire pipeline from the beginning, including the document parsing step that already succeeded. After reviewing the logs, you want to redesign how the field extractor handles and communicates this class of failure. Which approach best enables the coordinator to make intelligent, targeted recovery decisions?