{"id":18135,"date":"2026-09-22T05:37:43","date_gmt":"2026-09-22T05:37:43","guid":{"rendered":"https:\/\/www.examlabs.com\/certification\/?p=18135"},"modified":"2026-09-22T05:37:43","modified_gmt":"2026-09-22T05:37:43","slug":"databricks-certified-generative-ai-engineer-associate-practice-test-questions-and-exam-dumps-part7-q121-140","status":"publish","type":"post","link":"https:\/\/www.examlabs.com\/certification\/databricks-certified-generative-ai-engineer-associate-practice-test-questions-and-exam-dumps-part7-q121-140\/","title":{"rendered":"Databricks Certified Generative AI Engineer Associate Practice Test Questions and Exam Dumps Part7 Q121-140"},"content":{"rendered":"

View Full <\/b>Databricks Certified Generative AI Engineer Associate Exam Dumps<\/b><\/a> and Practice Test Dumps<\/b><\/p>\n

 <\/p>\n

Question 121. Which approach can improve retrieval when a user’s query contains ambiguous terminology?<\/b><\/p>\n

    \n
  1. <\/b> Remove all query terms<\/span>
    \n<\/span>2.<\/b> Use query rewriting or clarification<\/span>
    \n<\/span>3.<\/b> Disable semantic search<\/span>
    \n<\/span>4.<\/b> Increase response temperature<\/span><\/li>\n<\/ol>\n

    Correct Answer: 2. Use query rewriting or clarification<\/b><\/p>\n

    Explanation:<\/b><\/p>\n

    Ambiguous terminology can make it difficult for a retrieval system to identify the information the user actually needs. Query rewriting can transform an unclear question into a more precise search query using available conversational context or known terminology. In some situations, the application may instead ask the user for clarification before performing retrieval. These approaches help reduce irrelevant results and improve the likelihood that retrieved documents match the intended meaning. Increasing generation temperature does not resolve ambiguity in retrieval because temperature affects model output behavior. Similarly, removing query terms or disabling semantic search generally reduces the information available to the retrieval process.<\/span><\/p>\n

    Question 122. What is the purpose of storing metadata alongside document chunks in a RAG system?<\/b><\/p>\n

      \n
    1. <\/b> To provide filtering and additional contextual information<\/span>
      \n<\/span>2.<\/b> To replace the embedding model<\/span>
      \n<\/span>3.<\/b> To increase the model’s temperature<\/span>
      \n<\/span>4.<\/b> To eliminate document retrieval<\/span><\/li>\n<\/ol>\n

      Correct Answer: 1. To provide filtering and additional contextual information<\/b><\/p>\n

      Explanation:<\/b><\/p>\n

      Metadata provides information about a document or chunk beyond its textual content. Examples can include document title, source, author, department, creation date, language, product category, or access classification. This information can be used to filter retrieval results and ensure that only appropriate documents are considered. Metadata can also help applications display citations or identify where retrieved information originated. Metadata does not replace embeddings when semantic search is required, and it does not eliminate retrieval. Instead, it complements vector or keyword search by providing structured attributes that can improve relevance, governance, and traceability throughout a RAG workflow.<\/span><\/p>\n

      Question 123. Which characteristic is especially important when choosing a foundation model for a generative AI application?<\/b><\/p>\n

        \n
      1. <\/b> The application’s screen resolution<\/span>
        \n<\/span>2.<\/b> The model’s suitability for the required tasks and constraints<\/span>
        \n<\/span>3.<\/b> The number of browser extensions installed<\/span>
        \n<\/span>4.<\/b> The color of the dashboard<\/span><\/li>\n<\/ol>\n

        Correct Answer: 2. The model’s suitability for the required tasks and constraints<\/b><\/p>\n

        Explanation:<\/b><\/p>\n

        Foundation model selection should be based on the application’s functional and operational requirements. Important factors may include the model’s ability to follow instructions, language support, reasoning capabilities, context-window requirements, latency, cost, output quality, deployment options, and compatibility with the intended workload. Different models may perform differently across tasks such as summarization, question answering, structured generation, coding, or domain-specific assistance. A model should therefore be evaluated using representative application data rather than selected solely on unrelated characteristics. Testing candidate models against measurable requirements helps determine whether they are appropriate for the application’s intended use and operating environment.<\/span><\/p>\n

        Question 124. Why can a reranker improve RAG response quality?<\/b><\/p>\n

          \n
        1. <\/b> It selects or prioritizes the most relevant retrieved passages<\/span>
          \n<\/span>2.<\/b> It permanently retrains the language model<\/span>
          \n<\/span>3.<\/b> It removes all source documents<\/span>
          \n<\/span>4.<\/b> It disables vector search<\/span><\/li>\n<\/ol>\n

          Correct Answer: 1. It selects or prioritizes the most relevant retrieved passages<\/b><\/p>\n

          Explanation:<\/b><\/p>\n

          A reranker can improve RAG quality by evaluating an initial set of retrieved candidates more carefully and placing the most relevant passages higher in the final retrieval results. Initial vector or keyword search is designed to efficiently identify candidates, but similarity scores do not always perfectly represent query relevance. A reranker can use a more detailed comparison between the query and each candidate to refine the ordering. The resulting context can be more focused and useful for the generation model. Reranking does not retrain the language model, remove source documents, or disable vector search. It enhances the retrieval stage by improving candidate prioritization.<\/span><\/p>\n

          Question 125. What is one reason to use a similarity threshold during vector retrieval?<\/b><\/p>\n

            \n
          1. <\/b> To ensure that only results above a defined relevance level are considered<\/span>
            \n<\/span>2.<\/b> To guarantee every response is correct<\/span>
            \n<\/span>3.<\/b> To increase the model’s context window<\/span>
            \n<\/span>4.<\/b> To replace metadata<\/span><\/li>\n<\/ol>\n

            Correct Answer: 1. To ensure that only results above a defined relevance level are considered<\/b><\/p>\n

            Explanation:<\/b><\/p>\n

            A similarity threshold can prevent documents with very weak relationships to the user’s query from being included in the retrieved context. The retrieval system calculates a similarity or distance measure between the query embedding and candidate document embeddings, and the threshold determines which candidates are sufficiently close to be considered. This can reduce irrelevant context and help the application decide when available evidence may be insufficient. However, a threshold does not guarantee factual correctness because semantic similarity does not necessarily mean that a document contains a complete or accurate answer. Thresholds should therefore be evaluated and tuned using representative retrieval and application-quality data.<\/span><\/p>\n

            Question 126. Which practice can help prevent obsolete information from being returned by a RAG system?<\/b><\/p>\n

              \n
            1. <\/b> Continuously increase generation temperature<\/span>
              \n<\/span>2.<\/b> Maintain and update the indexed knowledge source<\/span>
              \n<\/span>3.<\/b> Remove all metadata<\/span>
              \n<\/span>4.<\/b> Disable document ingestion<\/span><\/li>\n<\/ol>\n

              Correct Answer: 2. Maintain and update the indexed knowledge source<\/b><\/p>\n

              Explanation:<\/b><\/p>\n

              A RAG application depends on the quality and freshness of the knowledge source used for retrieval. If documents are updated but the retrieval index continues to contain obsolete versions, the system may return outdated information to the generation model. Maintaining an ingestion and indexing process that detects updates, removes obsolete content, and refreshes relevant records can help keep retrieval results current. Metadata such as version or effective date can further support filtering and document lifecycle management. Increasing model temperature does not update knowledge, while disabling ingestion prevents new information from entering the retrieval system. Data freshness is therefore an important part of RAG maintenance.<\/span><\/p>\n

              Question 127. What is the primary role of an LLM in a typical RAG architecture?<\/b><\/p>\n

                \n
              1. <\/b> Generate a response using the supplied instructions and retrieved context<\/span>
                \n<\/span>2.<\/b> Permanently store every source document<\/span>
                \n<\/span>3.<\/b> Replace the vector index<\/span>
                \n<\/span>4.<\/b> Authenticate every user<\/span><\/li>\n<\/ol>\n

                Correct Answer: 1. Generate a response using the supplied instructions and retrieved context<\/b><\/p>\n

                Explanation:<\/b><\/p>\n

                In a typical RAG architecture, the language model receives a prompt containing instructions, the user’s question, and relevant information retrieved from external sources. It uses this supplied context to generate a natural-language response. The retrieval system and vector index are responsible for finding relevant information, while authentication and authorization are separate security functions. The language model does not generally serve as the permanent storage layer for source documents. Separating retrieval from generation allows the application to update external knowledge without necessarily changing model weights. This architecture can support grounded responses when the retrieved context is relevant, accurate, and appropriately incorporated into the prompt.<\/span><\/p>\n

                Question 128. Which design can help reduce hallucinations in a knowledge-based assistant?<\/b><\/p>\n

                  \n
                1. <\/b> Encourage the model to invent missing information<\/span>
                  \n<\/span>2.<\/b> Provide relevant retrieved evidence and instruct the model to stay grounded<\/span>
                  \n<\/span>3.<\/b> Remove source context<\/span>
                  \n<\/span>4.<\/b> Always maximize response length<\/span><\/li>\n<\/ol>\n

                  Correct Answer: 2. Provide relevant retrieved evidence and instruct the model to stay grounded<\/b><\/p>\n

                  Explanation:<\/b><\/p>\n

                  Providing relevant evidence gives the model information it can use when answering questions instead of relying entirely on its pretrained knowledge or generating unsupported content. A system instruction can further encourage the assistant to base its response on the supplied context and acknowledge when sufficient information is unavailable. Retrieval quality remains important because irrelevant or inaccurate evidence can still lead to poor answers. Maximizing response length does not improve factual grounding and may introduce additional unsupported statements. A combination of high-quality source data, effective retrieval, appropriate prompting, output controls, and evaluation can reduce hallucination risk, although no single technique guarantees that hallucinations will be eliminated.<\/span><\/p>\n

                  Question 129. What is an important consideration when designing prompts for production applications?<\/b><\/p>\n

                    \n
                  1. <\/b> Prompts should be consistent, testable, and appropriate for the intended task<\/span>
                    \n<\/span>2.<\/b> Prompts should contain every available document<\/span>
                    \n<\/span>3.<\/b> Prompts should always be as long as possible<\/span>
                    \n<\/span>4.<\/b> Prompts should never be versioned<\/span><\/li>\n<\/ol>\n

                    Correct Answer: 1. Prompts should be consistent, testable, and appropriate for the intended task<\/b><\/p>\n

                    Explanation:<\/b><\/p>\n

                    Production prompts should be designed deliberately and evaluated using representative examples. Clear instructions can help establish the expected task, output format, behavioral constraints, and use of retrieved information. Prompts should also be versioned so that changes can be tracked and compared during development. Including every available document can waste context capacity and introduce irrelevant information. Likewise, a longer prompt is not automatically a better prompt; unnecessary instructions can increase token usage and potentially create conflicting guidance. A production prompt should therefore balance clarity, specificity, maintainability, and context efficiency while being tested against realistic inputs and failure cases.<\/span><\/p>\n

                    Question 130. Which metric is most directly associated with the speed of an application’s response?<\/b><\/p>\n

                      \n
                    1. <\/b> Groundedness<\/span>
                      \n<\/span>2.<\/b> Retrieval precision<\/span>
                      \n<\/span>3.<\/b> Latency<\/span>
                      \n<\/span>4.<\/b> Answer relevance<\/span><\/li>\n<\/ol>\n

                      Correct Answer: 3. Latency<\/b><\/p>\n

                      Explanation:<\/b><\/p>\n

                      Latency measures the time required to process a request and produce a response, making it the metric most directly associated with response speed. In a generative AI application, teams may measure total latency as well as the latency of individual components such as retrieval, reranking, prompt construction, model inference, and downstream services. Understanding these components can help identify performance bottlenecks. Groundedness and answer relevance are quality-related measures, while retrieval precision evaluates the proportion of retrieved results that are relevant. Monitoring latency alongside quality and reliability metrics provides a more complete view of production performance and helps teams identify opportunities for optimization.<\/span><\/p>\n

                      Question 131. Why might an application use a smaller language model for a particular task?<\/b><\/p>\n

                        \n
                      1. <\/b> Smaller models can sometimes provide adequate quality with lower latency or cost<\/span>
                        \n<\/span>2.<\/b> Smaller models always have more knowledge<\/span>
                        \n<\/span>3.<\/b> Smaller models automatically eliminate hallucinations<\/span>
                        \n<\/span>4.<\/b> Smaller models never require evaluation<\/span><\/li>\n<\/ol>\n

                        Correct Answer: 1. Smaller models can sometimes provide adequate quality with lower latency or cost<\/b><\/p>\n

                        Explanation:<\/b><\/p>\n

                        A smaller language model may be appropriate when the application’s task does not require the capabilities of a larger model. Depending on the workload, a smaller model can provide lower inference costs, reduced latency, and lower resource requirements while still meeting quality expectations. However, model size alone does not determine suitability. The selected model should be evaluated against representative application tasks, required accuracy, language support, context needs, safety requirements, and operational constraints. Smaller models do not automatically eliminate hallucinations or guarantee quality. Model selection should therefore be based on measurable requirements and evaluation results rather than assuming that the largest available model is always necessary.<\/span><\/p>\n

                        Question 132. What is a key benefit of structured outputs from a generative AI model?<\/b><\/p>\n

                          \n
                        1. <\/b> They can make model responses easier for downstream software to process<\/span>
                          \n<\/span>2.<\/b> They eliminate the need for testing<\/span>
                          \n<\/span>3.<\/b> They permanently store generated responses<\/span>
                          \n<\/span>4.<\/b> They guarantee factual correctness<\/span><\/li>\n<\/ol>\n

                          Correct Answer: 1. They can make model responses easier for downstream software to process<\/b><\/p>\n

                          Explanation:<\/b><\/p>\n

                          Structured outputs allow a generative AI model to return information in a predictable format, such as JSON containing predefined fields. This can make integration with application code, databases, workflows, and APIs more reliable because downstream components can process expected fields instead of interpreting unrestricted natural language. Schema validation can further detect missing or malformed values before the output is consumed. Structured output does not guarantee that the information is factually correct, so semantic evaluation and grounding remain important. It also does not eliminate testing or serve as permanent storage. Its main benefit is improving consistency and interoperability between generative models and deterministic software components.<\/span><\/p>\n

                          Question 133. Which approach is appropriate for evaluating whether a model follows a required response format?<\/b><\/p>\n

                            \n
                          1. <\/b> Measure only network latency<\/span>
                            \n<\/span>2.<\/b> Validate the generated output against the required schema<\/span>
                            \n<\/span>3.<\/b> Remove output requirements<\/span>
                            \n<\/span>4.<\/b> Increase the number of retrieved documents<\/span><\/li>\n<\/ol>\n

                            Correct Answer: 2. Validate the generated output against the required schema<\/b><\/p>\n

                            Explanation:<\/b><\/p>\n

                            Schema validation provides a direct way to determine whether generated output follows a predefined structural requirement. For example, an application may require specific fields, data types, nested objects, or arrays. A validator can check these requirements automatically and identify responses that are incomplete or malformed. This is particularly useful when model output is consumed by downstream software. Network latency does not measure format compliance, and adding more retrieved documents does not ensure that the model will follow an output structure. Removing requirements makes evaluation impossible against a defined format. Schema validation can therefore be combined with semantic and factual evaluation for a more complete quality assessment.<\/span><\/p>\n

                            Question 134. What can metadata-based access filtering help prevent in an enterprise RAG application?<\/b><\/p>\n

                              \n
                            1. <\/b> Unauthorized users receiving restricted retrieved content<\/span>
                              \n<\/span>2.<\/b> All model latency<\/span>
                              \n<\/span>3.<\/b> Every possible hallucination<\/span>
                              \n<\/span>4.<\/b> Tokenization of input text<\/span><\/li>\n<\/ol>\n

                              Correct Answer: 1. Unauthorized users receiving restricted retrieved content<\/b><\/p>\n

                              Explanation:<\/b><\/p>\n

                              Metadata-based access filtering can help restrict retrieval to documents that a particular user or service is authorized to access. For example, documents may contain metadata identifying an organization, department, security classification, or access group. The retrieval layer can use these attributes to exclude records that should not be available to the requesting user. This is an important consideration for enterprise RAG systems because the model may otherwise receive restricted information and potentially include it in a response. Access filtering does not eliminate all hallucinations or latency, and it is not related to tokenization. It should also be supported by broader authentication and authorization controls.<\/span><\/p>\n

                              Question 135. Why is human evaluation sometimes included in generative AI testing?<\/b><\/p>\n

                                \n
                              1. <\/b> Human reviewers can assess qualities that automated metrics may not fully capture<\/span>
                                \n<\/span>2.<\/b> Humans automatically improve model weights<\/span>
                                \n<\/span>3.<\/b> Human evaluation eliminates the need for automated tests<\/span>
                                \n<\/span>4.<\/b> Human reviewers replace the retrieval system<\/span><\/li>\n<\/ol>\n

                                Correct Answer: 1. Human reviewers can assess qualities that automated metrics may not fully capture<\/b><\/p>\n

                                Explanation:<\/b><\/p>\n

                                Human evaluation can provide valuable insight into qualities that may be difficult to measure reliably using automated metrics alone. Reviewers can assess whether responses are useful, clear, appropriate, coherent, sufficiently grounded, and aligned with user expectations. This can be particularly important for subjective or domain-specific tasks where simple similarity metrics may not capture meaningful differences between responses. Human evaluation does not replace automated testing because automated evaluation enables scalable and repeatable comparisons across large datasets. Instead, human feedback can complement automated metrics and help identify failure patterns or quality dimensions that require deeper interpretation during model and application evaluation.<\/span><\/p>\n

                                Question 136. What should happen before deploying a significant model or prompt change to production?<\/b><\/p>\n

                                  \n
                                1. <\/b> The change should be evaluated against representative tests<\/span>
                                  \n<\/span>2.<\/b> All monitoring should be disabled<\/span>
                                  \n<\/span>3.<\/b> Existing evaluation data should be deleted<\/span>
                                  \n<\/span>4.<\/b> The change should be deployed without comparison<\/span><\/li>\n<\/ol>\n

                                  Correct Answer: 1. The change should be evaluated against representative tests<\/b><\/p>\n

                                  Explanation:<\/b><\/p>\n

                                  Significant changes to a generative AI application should be evaluated before production deployment to identify potential improvements and regressions. Representative tests can include known questions, difficult edge cases, safety scenarios, retrieval tests, and other examples that reflect expected production usage. Comparing the new configuration with a baseline can show whether the change affects quality, latency, cost, or other important metrics. Deploying without evaluation makes it harder to detect regressions before users encounter them. Monitoring should remain available after deployment as an additional safeguard. Controlled evaluation and staged deployment together support safer iteration and better visibility into application behavior.<\/span><\/p>\n

                                  Question 137. What is a benefit of using a representative evaluation dataset?<\/b><\/p>\n

                                    \n
                                  1. <\/b> It can provide a more realistic indication of application performance<\/span>
                                    \n<\/span>2.<\/b> It guarantees perfect production behavior<\/span>
                                    \n<\/span>3.<\/b> It removes the need for monitoring<\/span>
                                    \n<\/span>4.<\/b> It prevents every security issue<\/span><\/li>\n<\/ol>\n

                                    Correct Answer: 1. It can provide a more realistic indication of application performance<\/b><\/p>\n

                                    Explanation:<\/b><\/p>\n

                                    A representative evaluation dataset reflects the types of questions, documents, user behaviors, and edge cases that an application is expected to encounter. Testing against such data provides a more meaningful indication of how different configurations may perform in realistic scenarios. If the dataset contains only simple or unusual examples, evaluation results may not generalize well to production usage. A representative dataset does not guarantee perfect behavior because real-world conditions can differ and new failure cases can appear. It also does not replace monitoring or security controls. Its primary purpose is to make comparative evaluation more relevant and informative.<\/span><\/p>\n

                                    Question 138. Which technique can help reduce the number of irrelevant retrieved documents before generation?<\/b><\/p>\n

                                      \n
                                    1. <\/b> Reranking and relevance filtering<\/span>
                                      \n<\/span>2.<\/b> Increasing output randomness<\/span>
                                      \n<\/span>3.<\/b> Removing all retrieval scores<\/span>
                                      \n<\/span>4.<\/b> Increasing conversation history indefinitely<\/span><\/li>\n<\/ol>\n

                                      Correct Answer: 1. Reranking and relevance filtering<\/b><\/p>\n

                                      Explanation:<\/b><\/p>\n

                                      Reranking and relevance filtering can refine the results returned by an initial retrieval operation. The retrieval system may first identify a set of candidate documents, after which a reranker can evaluate their relevance more carefully. A similarity threshold or other filtering mechanism can then exclude candidates that do not meet the application’s relevance requirements. Reducing irrelevant context is useful because it allows the generation model to focus on stronger evidence and can reduce unnecessary token consumption. Increasing output randomness does not improve retrieval quality, while removing retrieval scores eliminates useful ranking information. Unlimited conversation history can also introduce additional irrelevant context rather than improving document relevance.<\/span><\/p>\n

                                      Question 139. What is an important reason to retain source information for retrieved passages?<\/b><\/p>\n

                                        \n
                                      1. <\/b> It can support traceability and help users understand where information originated<\/span>
                                        \n<\/span>2.<\/b> It increases model temperature<\/span>
                                        \n<\/span>3.<\/b> It eliminates the need for embeddings<\/span>
                                        \n<\/span>4.<\/b> It guarantees every source is accurate<\/span><\/li>\n<\/ol>\n

                                        Correct Answer: 1. It can support traceability and help users understand where information originated<\/b><\/p>\n

                                        Explanation:<\/b><\/p>\n

                                        Retaining source information for retrieved passages supports traceability by allowing an application to identify where the information used in a response came from. Depending on the design, this may include document identifiers, titles, URLs, timestamps, or other source metadata. Showing appropriate source references can help users verify important information and can help developers investigate incorrect responses. Source tracking does not guarantee that the underlying source is accurate, so source quality and evaluation remain necessary. It also does not eliminate the need for embeddings or affect model temperature. Traceability is an important part of building transparent and maintainable RAG applications.<\/span><\/p>\n

                                        Question 140. Which overall architecture best supports a grounded enterprise knowledge assistant?<\/b><\/p>\n

                                          \n
                                        1. <\/b> User query \u2192 model generates from memory only<\/span>
                                          \n<\/span>2.<\/b> User query \u2192 retrieve authorized relevant context \u2192 generate with instructions \u2192 validate and monitor<\/span>
                                          \n<\/span>3.<\/b> User query \u2192 remove source documents \u2192 generate unrestricted output<\/span>
                                          \n<\/span>4.<\/b> User query \u2192 randomly change the model \u2192 return response<\/span><\/li>\n<\/ol>\n

                                          Correct Answer: 2. User query \u2192 retrieve authorized relevant context \u2192 generate with instructions \u2192 validate and monitor<\/b><\/p>\n

                                          Explanation:<\/b><\/p>\n

                                          A grounded enterprise knowledge assistant typically combines several components rather than relying on the language model alone. The user’s query can be processed and used to retrieve relevant information from an authorized knowledge source. The retrieved context is then provided to the model together with appropriate system instructions. Output validation can enforce required formats or other constraints, while monitoring provides visibility into quality, latency, errors, and usage after deployment. Authorization and retrieval controls are particularly important when enterprise data contains restricted information. This architecture separates knowledge retrieval, generation, validation, and operational monitoring, creating a more structured foundation for reliable generative AI applications.<\/span><\/p>\n

                                           <\/p>\n","protected":false},"excerpt":{"rendered":"

                                          View Full Databricks Certified Generative AI Engineer Associate Exam Dumps and Practice Test Dumps   Question 121. Which approach can improve retrieval when a user’s query contains ambiguous terminology? Remove all query terms 2. Use query rewriting or clarification 3. Disable semantic search 4. Increase response temperature Correct Answer: 2. Use query rewriting or clarification […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[1648,1647],"tags":[],"_links":{"self":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts\/18135"}],"collection":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/comments?post=18135"}],"version-history":[{"count":1,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts\/18135\/revisions"}],"predecessor-version":[{"id":18136,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/posts\/18135\/revisions\/18136"}],"wp:attachment":[{"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/media?parent=18135"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/categories?post=18135"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.examlabs.com\/certification\/wp-json\/wp\/v2\/tags?post=18135"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}