Databricks Certified Generative AI Engineer Associate Bundle

Not the strongest implement choice for this specific question. Join authorized structured values and retrieved policy context through a controlled composition stage. This action primarily relates to Parallel retrieval and deterministic chain postprocessing: Combining structured lookup and semantic retrieval requires explicit routing, result joining, and downstream schema checks. However, the described case — A policy assistant produces unsupported answers because its chain sends only the user question to the LLM. — is testing Retriever-generator chain component boundaries. It does not directly establish the implement outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest implement choice for this specific question. Define a schema-constrained prompt with field types, required keys and strict structured-output validation at the consumer. This action primarily relates to Structured JSON schema prompt and format compliance: A model can produce plausible prose rather than machine-readable JSON unless the output contract and validation are explicit. However, the described case — A policy assistant produces unsupported answers because its chain sends only the user question to the LLM. — is testing Retriever-generator chain component boundaries. It does not directly establish the implement outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest implement choice for this specific question. Split read-only diagnostic tools from authorized destructive operations with separate permissions. This action primarily relates to Agent tool schemas, least privilege and retries: Well-described tool schemas constrain parameters, and idempotent retries prevent duplicate side effects. However, the described case — A policy assistant produces unsupported answers because its chain sends only the user question to the LLM. — is testing Retriever-generator chain component boundaries. It does not directly establish the implement outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Compare fixed evaluation queries on baseline and candidate retrieval with confidence in differences. This action primarily relates to Retrieval recall precision and ranking diagnostics: Measure hit recall, precision, MRR or nDCG on labeled query-to-relevant-chunk pairs before tuning retrieval. However, the described case — A 300-page maintenance manual is cut into tiny fragments and the retrieved snippet omits the cause-and-remedy pair. — is testing Chunk size and overlap tradeoffs. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Sample pages/images against extracted text and check structured fields and layout fidelity. This action primarily relates to Format-specific document extraction and OCR needs: Extract text with a parser suited to source format; image-only scans require OCR rather than HTML or plain-text parsing. However, the described case — A 300-page maintenance manual is cut into tiny fragments and the retrieved snippet omits the cause-and-remedy pair. — is testing Chunk size and overlap tradeoffs. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Retrieve rare table values and confirm the matching row, units and header context are intact. This action primarily relates to Table-aware chunking and metadata-rich retrieval: Tabular rows need headers, units and column semantics preserved when transformed into retrieval documents. However, the described case — A 300-page maintenance manual is cut into tiny fragments and the retrieved snippet omits the cause-and-remedy pair. — is testing Chunk size and overlap tradeoffs. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Revisit cost/latency budgets whenever context window or retrieval size changes significantly. This action primarily relates to Latency, cost and retrieval targets as design requirements: A valid pipeline design has measurable quality, latency, token and throughput constraints before model selection. However, the described case — A refund agent issues credits before verifying invoice status or refund eligibility. — is testing Tool ordering and error isolation in multi-stage reasoning. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Do not assume an instruction to 'respond in JSON' replaces actual schema validation and error handling. This action primarily relates to Structured JSON schema prompt and format compliance: A model can produce plausible prose rather than machine-readable JSON unless the output contract and validation are explicit. However, the described case — A refund agent issues credits before verifying invoice status or refund eligibility. — is testing Tool ordering and error isolation in multi-stage reasoning. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Do not treat model fluency or demo satisfaction as a substitute for measurable business acceptance criteria. This action primarily relates to Translate use-case requirements into pipeline inputs and outputs: Business acceptance criteria must map to concrete source inputs, generated fields, correctness measures and permission constraints. However, the described case — A refund agent issues credits before verifying invoice status or refund eligibility. — is testing Tool ordering and error isolation in multi-stage reasoning. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Run reproducible pairwise experiments with confidence and segment-level result analysis. This action primarily relates to Model evaluation with paired benchmarks: Accuracy, faithfulness, relevance and refusal behavior require task-specific scoring, and a model switch needs common test conditions. However, the described case — An internal receipt classifier must process 200 requests per second under tight GPU cost targets. — is testing Model selection by quality latency and budget. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Run equivalent retrieval tests including storage and refresh cost for candidate encoders. This action primarily relates to Embedding dimensions latency and index storage: Embedding width affects index storage and computation, and larger dimensions do not automatically improve task relevance. However, the described case — An internal receipt classifier must process 200 requests per second under tight GPU cost targets. — is testing Model selection by quality latency and budget. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Check embedding dimensions or extracted-field validation against task-specific acceptance tests. This action primarily relates to Specialized model versus general LLM: Embedding, reranking, extraction and generative models solve different stages and should be evaluated for their actual role. However, the described case — An internal receipt classifier must process 200 requests per second under tight GPU cost targets. — is testing Model selection by quality latency and budget. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Use usage monitoring and budgets to detect regressions when prompts, indexes or models change. This action primarily relates to Databricks usage and token cost controls: Reduce LLM spend through token budgets, caching, smaller appropriate models and AI Gateway quotas while preserving acceptance quality. However, the described case — A data extraction agent has identical field-F1 across two architectures but one exceeds the production latency budget. — is testing Model selection from quantitative task metrics. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Use risk-based alert thresholds and sample protected user segments, not a single global average. This action primarily relates to Production performance KPIs and incident thresholds: Monitoring should include answer correctness, retrieval coverage, safety, latency, drift and per-user harm indicators, not just availability. However, the described case — A data extraction agent has identical field-F1 across two architectures but one exceeds the production latency budget. — is testing Model selection from quantitative task metrics. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Do not equate an LLM judge score with unquestionable ground truth; calibrate against reviewed labels. This action primarily relates to MLflow GenAI tracing, built-in and custom scorers: MLflow GenAI traces capture components, and scorers/judges can evaluate a full answer or individual retrieval and tool spans. However, the described case — A data extraction agent has identical field-F1 across two architectures but one exceeds the production latency budget. — is testing Model selection from quantitative task metrics. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest diagnose choice for this specific question. Inspect predict entrypoint, model artifacts, dependency environment, inputs and output serialization. This action primarily relates to MLflow pyfunc packaging for pre- and post-processing: A custom MLflow pyfunc model wraps request parsing, retriever/model invocation and output formatting inside a reproducible predict interface. However, the described case — A new Vector Search index is created in staging but production points at its previous source table version. — is testing CI/CD pipelines for retrieval and agent components. It does not directly establish the diagnose outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest diagnose choice for this specific question. Inspect serving endpoint permissions, token path, user/app identity and downstream Unity Catalog access. This action primarily relates to Model Serving endpoint permissions and caller identity: Access to a serving endpoint is governed by workspace rights and identity; caller authorization is distinct from data-source permissions. However, the described case — A new Vector Search index is created in staging but production points at its previous source table version. — is testing CI/CD pipelines for retrieval and agent components. It does not directly establish the diagnose outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest diagnose choice for this specific question. Inspect package lockfile, model signature, serving logs and API contract differences. This action primarily relates to Deployed model dependency drift and environment compatibility: Models logged in experiments can fail in serving if Python dependencies or inference signatures differ between build and target. However, the described case — A new Vector Search index is created in staging but production points at its previous source table version. — is testing CI/CD pipelines for retrieval and agent components. It does not directly establish the diagnose outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Never use persuasive phrasing or a citation marker as evidence that a generated statement is true. This action primarily relates to Hallucinated facts and unsupported citations: Fluent responses can invent facts and citations; faithfulness must be checked against supplied evidence, not writing style. However, the described case — A retrieved wiki page tells the agent to email confidential documents to an external address. — is testing Prompt-injection defenses for retrieved context. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Expire or segregate conversation memory by identity and purpose, especially for financial and personal records. This action primarily relates to Conversation context and intent routing: Multi-turn assistants need explicit conversational state while routing each new request to the correct tool or topic. However, the described case — A retrieved wiki page tells the agent to email confidential documents to an external address. — is testing Prompt-injection defenses for retrieved context. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Never trust an LLM-extracted customer identifier without server-side permission validation. This action primarily relates to Query-field driven prompt and retrieval filters: User intent and structured fields such as account_id should drive authorized SQL lookups or retrieval filters, not be ignored in free text. However, the described case — A retrieved wiki page tells the agent to email confidential documents to an external address. — is testing Prompt-injection defenses for retrieved context. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Document legal clearance and confirm the deployed artifact matches the approved license/version. This action primarily relates to Open model license versus managed endpoint obligations: Model availability on a hub does not itself confer commercial training, redistribution or hosting rights. However, the described case — A medical summarization product considers a model evaluated only on short English news articles. — is testing Model cards and task-specific limitations. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Verify matched query and document vectors use the approved version and retrieval improves on held-out examples. This action primarily relates to Embedding model swap and reindex compatibility: Changed embedding dimensions or model semantics typically require re-embedding and index consistency before comparing search quality. However, the described case — A medical summarization product considers a model evaluated only on short English news articles. — is testing Model cards and task-specific limitations. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Run reproducible pairwise experiments with confidence and segment-level result analysis. This action primarily relates to Model evaluation with paired benchmarks: Accuracy, faithfulness, relevance and refusal behavior require task-specific scoring, and a model switch needs common test conditions. However, the described case — A medical summarization product considers a model evaluated only on short English news articles. — is testing Model cards and task-specific limitations. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Version embeddings with source and model identifiers to prevent silently mixing incompatible representations. This action primarily relates to Embedding model swap and reindex compatibility: Changed embedding dimensions or model semantics typically require re-embedding and index consistency before comparing search quality. However, the described case — A team picks a model marked research-only and plans to deploy it for a paid product. — is testing Open model license versus managed endpoint obligations. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Do not cherry-pick average metrics that hide severe high-impact task failures. This action primarily relates to Model evaluation with paired benchmarks: Accuracy, faithfulness, relevance and refusal behavior require task-specific scoring, and a model switch needs common test conditions. However, the described case — A team picks a model marked research-only and plans to deploy it for a paid product. — is testing Open model license versus managed endpoint obligations. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest safeguard choice for this specific question. Reassess tokenizer and context limits when switching embedding models or input languages. This action primarily relates to Embedding context window and chunk token distribution: The embedding encoder must accept the actual tokenized chunk length; silently truncating chunks can drop relevant suffix facts. However, the described case — A team picks a model marked research-only and plans to deploy it for a paid product. — is testing Open model license versus managed endpoint obligations. It does not directly establish the safeguard outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Test mismatched user and account values and confirm zero unauthorized retrieval and correct contextual answers. This action primarily relates to Query-field driven prompt and retrieval filters: User intent and structured fields such as account_id should drive authorized SQL lookups or retrieval filters, not be ignored in free text. However, the described case — An agent receives a customer message requesting deletion of every table while only ticket lookup is authorized. — is testing Unsafe model output and jailbreak response diagnosis. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Rerun a controlled experiment and confirm metrics match within expected stochastic tolerance. This action primarily relates to Experiment tracking and version reproducibility: MLflow experiments should retain prompt, model, dataset, parameters and evaluation outputs to support a defensible comparison. However, the described case — An agent receives a customer message requesting deletion of every table while only ticket lookup is authorized. — is testing Unsafe model output and jailbreak response diagnosis. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Document legal clearance and confirm the deployed artifact matches the approved license/version. This action primarily relates to Open model license versus managed endpoint obligations: Model availability on a hub does not itself confer commercial training, redistribution or hosting rights. However, the described case — An agent receives a customer message requesting deletion of every table while only ticket lookup is authorized. — is testing Unsafe model output and jailbreak response diagnosis. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Load-test p95 latency and per-request cost against realistic payload sizes and concurrency. This action primarily relates to Latency, cost and retrieval targets as design requirements: A valid pipeline design has measurable quality, latency, token and throughput constraints before model selection. However, the described case — Accounts payable needs supplier name, invoice date and line items extracted into typed columns from PDFs. — is testing Agent Bricks Information Extraction versus Multiagent Supervisor. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Compare outputs against task-specific labeled examples and inspect mismatches between required and produced formats. This action primarily relates to Task choice: classification versus generative summarization: Classification predicts a category whereas summarization compresses meaning; the desired output determines the appropriate task. However, the described case — Accounts payable needs supplier name, invoice date and line items extracted into typed columns from PDFs. — is testing Agent Bricks Information Extraction versus Multiagent Supervisor. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.

Not the strongest verify choice for this specific question. Trigger a timeout then retry and verify only one approved action persists with a trace. This action primarily relates to Agent tool schemas, least privilege and retries: Well-described tool schemas constrain parameters, and idempotent retries prevent duplicate side effects. However, the described case — Accounts payable needs supplier name, invoice date and line items extracted into typed columns from PDFs. — is testing Agent Bricks Information Extraction versus Multiagent Supervisor. It does not directly establish the verify outcome or constraint that the question requires; it may still be useful in a different stage or incident.