AI Data Extraction Pipelines Quiz

Answers and explanations appear as you take the quiz.

1. What is the main goal of an AI data extraction pipeline?
   • To turn unstructured sources into clean, structured data fitting a schema
   • To generate marketing images
   • To translate speech to music
   • To compress video files
2. What does schema-constrained generation aim to control in an extraction pipeline?
   • The truth of every extracted value without review
   • The requested fields and output structure, while semantic correctness still requires validation
   • The original layout of a scanned document after OCR
   • Whether the source document is authentic
3. What is the role of an OCR or layout-parsing stage?
   • To delete the original file
   • To encrypt the document
   • To recover text and preserve table or form structure before extraction
   • To play the document as audio
4. How do well-designed pipelines handle low-confidence extractions?
   • They silently discard them
   • They always trust them anyway
   • They restart the whole pipeline from scratch
   • They route them to a human for review
5. What is one example of a validation rule in such a pipeline?
   • Line-item totals must add up and dates must be valid
   • The file must be blue
   • The model must use exactly 100 words
   • The document must be in landscape mode
6. What two main phases does the ChatSchema paper describe for structured medical-report extraction?
   • Training a speech recognizer and then translating subtitles
   • Classifying report scenarios and then extracting structured fields
   • Rendering a 3D scene and then estimating camera motion
   • Ranking product recommendations and then choosing a route
7. Why validate extracted fields against a schema?
   • To prove all extracted facts are true
   • To replace the source document
   • To catch missing, malformed, or out-of-range values before downstream use
   • To eliminate OCR errors
8. What provenance should a robust extraction pipeline retain?
   • Only the final JSON output
   • The model temperature only
   • A summary with no source reference
   • Links to the source document and location supporting each extracted field