Home / GenAI Engineer practice test / Design Applications / Question

Databricks Generative AI Engineer Associate · free practice question

A legal practice ingests digital contract documents (PDFs) and seeks to automate the population of its case management system. The goal is to extract key…

Exam-style question on Design Applications from the free GenAI Engineer practice test, with the answer and a full explanation below.

Updated November 11, 2026

Design Applications

A legal practice ingests digital contract documents (PDFs) and seeks to automate the population of its case management system. The goal is to extract key details such as party_names (all signatories), effective_date, governing_law_jurisdiction, total_contract_value, and payment_terms into the database. Manual processing currently handles 300 contracts weekly, each taking about 20 minutes. What Natural Language Processing (NLP) task type accurately describes the automation required for this workflow?

  1. AStructured data extraction: give the model a schema of the target fields and have it return them from the contract text as JSON
  2. BConversational question answering: ask a contract chatbot to describe each agreement and take the fields from its reply
  3. CDocument summarization: condense each contract into a short two-paragraph synopsis from which the key values are then read into the database
  4. DText generation: prompt an LLM to write the contract's metadata from its general knowledge of how such agreements are drafted
  5. ESemantic similarity search: embed each contract, find the closest match in a reference set, and reuse that contract's metadata
Answer & explanation

Correct answer: A

WHY A is correct: The requirement is to extract specific, named data fields from unstructured contract text and write them into a structured database — this is the definition of information extraction (also called structured output generation or named entity extraction with schema). The correct approach is to prompt a capable LLM with: (1) the contract text, (2) a schema defining the target fields, and (3) an instruction to return results as JSON. A foundation model served on Databricks with structured outputs (a JSON schema in response_format), or the ai_extract and ai_query SQL functions, can do this at scale and cut per-contract processing time from 20 minutes to seconds.

WHY NOT C: Summarization produces prose that condenses the document's content but does not extract specific fields into machine-readable structures. A human would still need to read the summary to find and copy each value — reducing but not eliminating manual effort. The goal is automated database population, not reading assistance.

WHY NOT B: Conversational chatbots produce unstructured natural language responses not formatted for database ingestion. Parsing a chatbot's conversational reply to extract structured field values adds significant fragility. A direct, schema-prompted extraction call is simpler, faster, and more reliable.

WHY NOT D: Text generation models do not 'know' the actual content of a specific uploaded contract — they have no access to document-specific information from training weights. Any metadata generated without reading the contract would be hallucinated. The model must receive the contract content in context.

WHY NOT E: Semantic similarity search identifies documents that are semantically similar to a reference, but two contracts being similar in topic does not mean they share the same parties, dates, or values. Copying metadata from a similar contract would produce incorrect results. Each contract requires its own extraction.

Take the full GenAI Engineer practice test free →

Would you pass today?

This question is one of 45 in the free, timed Databricks Generative AI Engineer Associate practice test. Take it and get a section-by-section score.

Start the free test →