Developer Releases PDF Freshness Checker Using LandingAI ADE and Parallel Decisions API

A developer released a document agent that checks time-sensitive claims in technical PDFs against live web evidence. The tool uses LandingAI Agentic Document Extraction to parse PDFs and Parallel Decisions API to classify claims as valid, outdated, or needing review. A test on an old OpenAI API Beta document extracted 12 claims: 7 outdated, 1 valid, and 4 needing review.
Event version 1.
Key insights
Targeted claim extraction
The agent identifies only freshness-sensitive claims such as versions, APIs, packages, configurations, deprecations, models, dependencies, and integrations rather than verifying every sentence. It preserves each claim's original statement and surrounding document context.
Three-state verification
Each claim is sent to Parallel Decisions API, which searches the live web and returns a bounded decision of VALID, OUTDATED, or NEEDS REVIEW with confidence and citations. Evidence that is insufficient or contradictory is classified as NEEDS REVIEW rather than forced into a verdict.
Dual evidence trails
Every result retains the exact excerpt from the uploaded document alongside live-web citations returned by Parallel. The output keeps the original claim and the current web evidence together for human judgment.
OpenAI API Beta test results
A test on an old OpenAI API Beta document extracted 12 claims: 7 were outdated, 1 remained valid, and 4 needed review. Outdated items included the old Engines route, davinci examples, and historical pricing.