AI0-001 Practice Test 28 — 15 Questions

Question 1

A developer is building an AI agent that needs to call external tools (e.g., weather API, database) and reason about the results to answer user queries. Which THREE components are essential for implementing this agentic workflow?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Planning capability (e.g., step-by-step decomposition). Option A (Planning capability) is correct because an agentic workflow requires the model to decompose a complex user query into an ordered sequence of steps, deciding which sub-tasks to execute and in what order before invoking tools. Option B (ReAct loop) is correct because the Reasoning + Acting pattern interleaves thought, action, and observation cycles, letting the agent call a tool, reason over the returned result, and decide the next action until the query is resolved. Option E (Function calling or tool use interface) is correct because the agent must have a structured mechanism (e.g., JSON schema-based function/tool definitions) to invoke the weather API or database and receive machine-readable outputs. Option C (Fine-tuned domain-specific model) is not essential, since a general-purpose LLM with tool-calling and reasoning prompts can drive the workflow without domain fine-tuning. Option D (Vector store for long-term memory) is not essential either, as retrieval-based long-term memory is an optional enhancement rather than a required component for calling tools and reasoning over their results.

Answer

Fine-tuned domain-specific model

Answer

A vector store for long-term memory

Question 2

A developer is building an AI microservice that processes document intelligence requests asynchronously. Users upload PDFs, and the service extracts text and analyzes it with an LLM. The processing time per document can be up to 5 minutes. Which integration pattern is MOST appropriate?

Accepted Answer

Async processing with a message queue and separate worker service. Async processing with a message queue and separate worker service is the correct pattern because document processing can take up to 5 minutes, which exceeds typical synchronous HTTP timeout limits (often 30-60 seconds). The API accepts the upload, enqueues a job, and returns immediately; a worker service consumes the queue, performs text extraction and LLM analysis, and stores results for later retrieval. This decouples request handling from long-running work and provides resilience against worker failures.

Answer

Synchronous REST API call that waits for the LLM response

Answer

WebSocket connection for real-time streaming

Answer

Serverless function triggered by HTTP request

Question 3

A data science team is preparing a dataset for a supervised learning task. They split the data into training and test sets. The team then normalizes the features using the mean and standard deviation calculated from the entire dataset before splitting. What issue does this introduce?

Accepted Answer

It introduces train/test leakage. Computing normalization statistics (mean and standard deviation) on the entire dataset before splitting means the test set's statistics influence the training data transformation. This is a form of train/test leakage: information from the test set leaks into the training pipeline, producing an optimistically biased estimate of model performance. The correct approach is to fit the scaler on the training set only and apply the same transformation to the test set.

Answer

It improves model generalization

Answer

It causes the model to overfit the training data

Answer

It reduces the variance of the features

Question 4

A chatbot application uses a system prompt to set the assistant's behavior. The developer wants the LLM to output structured JSON for downstream processing. Which technique BEST ensures the output is valid JSON?

Accepted Answer

Use the LLM's built-in JSON mode (e.g., response_format='json_object'). Many LLMs support a JSON mode that constrains output to valid JSON. System prompts can request JSON but may be ignored. Few-shot examples help but are not foolproof. Chain-of-thought is for reasoning, not formatting.

Answer

Set the temperature to 0 to make output deterministic

Answer

Include a few-shot example showing a JSON output in the prompt

Answer

Add a chain-of-thought reasoning step before the output

Question 5

Which TWO statements correctly describe the difference between supervised and unsupervised learning?

Accepted Answer

Supervised learning requires labeled data. Supervised learning relies on labeled datasets where each training example is paired with an output label, enabling the model to learn a mapping from inputs to outputs. This is a fundamental distinction from unsupervised learning, which works with unlabeled data to find inherent structures or patterns.

Answer

Supervised learning is only used for classification

Answer

Unsupervised learning always requires a target variable

Answer

Supervised learning is a subset of reinforcement learning

Question 6

An AI team notices that their hiring model consistently selects male candidates over equally qualified female candidates. Analysis shows the training data contains past hiring decisions where men were predominantly hired. Which type of bias is the root cause?

Accepted Answer

Historical bias. Historical bias is the root cause because the training data reflects past hiring decisions that systematically favored male candidates, encoding societal or organizational prejudices into the model. The model learns these historical patterns and perpetuates them, leading to discriminatory outcomes against equally qualified female candidates. This is distinct from algorithmic bias, which would arise from the model's design or optimization process itself.

Answer

Algorithmic bias

Answer

Confirmation bias

Answer

Selection bias

Question 7

The exhibit shows the output of a drift monitoring command for a fraud detection model. The team has an automated pipeline that triggers retraining when the overall average drift score exceeds 0.10. Based on the exhibit, what should the operations team do next?

Accepted Answer

Manually analyze the drift in 'amount' and 'location' and investigate potential causes.. The exhibit shows that the overall average drift score is below 0.10, so the automated retraining pipeline should not trigger. However, individual features like 'amount' and 'location' show elevated drift values that warrant manual investigation to understand root causes before any retraining decision. The team should analyze these specific features to determine if the drift is due to genuine data distribution changes or data quality issues.

Answer

Force retraining on all features to ensure the model adapts to the new data distribution.

Answer

No action is needed because the model is performing within acceptable drift limits.

Answer

Initiate the automated retraining pipeline since the average drift exceeds 0.05.

Question 8

Which of the following is a key characteristic of Narrow AI (Weak AI)?

Accepted Answer

It is designed to excel at a single, specific task. Narrow AI, also known as Weak AI, is designed and trained to perform a single, specific task with high proficiency, such as language translation, image recognition, or playing chess. It cannot generalize its intelligence to other domains, which distinguishes it from Artificial General Intelligence (AGI). Option C correctly captures this fundamental characteristic.

Answer

It can perform any intellectual task that a human can

Answer

It requires no training data

Answer

It surpasses human intelligence in all domains

Question 9

A company is forming an AI ethics board to oversee the development of a high-stakes AI system for bail decision recommendations. Which THREE responsibilities should the board primarily undertake?

Accepted Answer

Review model outputs for disparate impact across demographic groups. Option A is correct because an AI ethics board overseeing a bail recommendation system must audit model outputs for disparate impact across demographic groups, since bail decisions are legally and ethically sensitive and bias can violate anti-discrimination requirements. Option C is correct because the board should establish human-in-the-loop requirements for high-risk decisions, ensuring that consequential bail recommendations are reviewed by a qualified human rather than fully automated. Option D is correct because the board must define fairness criteria and acceptable bias thresholds, giving the organization measurable standards for evaluating whether the system's outputs are equitable. Option B is not a primary ethics-board responsibility because marketing the system to clients is a commercial function, not ethical oversight. Option E is not appropriate because writing production code is an engineering task, and the board should provide governance, review, and policy direction rather than implementation work.

Answer

Market the AI system to potential clients

Answer

Write the production code for the AI model

Question 10

A data engineer is designing a feature store for machine learning. Which THREE components are essential for a feature store? (Choose THREE.)

Accepted Answer

Data ingestion pipeline. A feature store fundamentally requires a data ingestion pipeline (A) to pull raw data from sources such as batch files, streaming events, or databases and transform it into features, since without ingestion there would be no features to store or serve. The online serving layer (B) is essential because it provides low-latency access to the latest feature values for real-time inference, typically backed by a low-latency store like Redis or DynamoDB. The feature repository (C) is also essential as the central catalog that stores feature definitions, metadata, and versioned feature values, enabling reuse and consistency between training and serving. Experiment tracking (D) and model registry (E) are important MLOps components but belong to the model development and deployment lifecycle, not to the core architecture of a feature store, so they are not essential components of a feature store itself.

Answer

Experiment tracking

Answer

Model registry

Question 11

A team is deploying a sentiment analysis model that must achieve high precision and high recall. They have a labeled dataset of 10,000 samples. They want to minimize overfitting. Which THREE actions are most appropriate? (Select THREE.)

Accepted Answer

Apply L2 regularization to the model weights. Option B is correct because applying L2 regularization adds a penalty proportional to the squared magnitude of the model weights to the loss function, which constrains weight growth and directly reduces overfitting, helping the model generalize and maintain both precision and recall on unseen data. Option C is correct because dropout layers randomly deactivate a fraction of neurons during each training iteration, preventing the network from relying on specific co-adapted units and acting as an effective regularizer that lowers overfitting. Option E is correct because augmenting the 10,000-sample training set with synthetic examples increases data diversity and effective sample size, which is a standard technique to improve generalization and reduce overfitting when labeled data is limited. Option A is not appropriate because decreasing the learning rate only affects optimization step size and convergence stability, not the model's capacity to overfit, and can even slow convergence without regularizing. Option D is not appropriate because increasing the training batch size changes gradient estimation variance and training dynamics but does not by itself prevent overfitting; in fact, very large batches can sometimes harm generalization.

Answer

Decrease the learning rate

Answer

Increase the training batch size

Question 12

A data science team is deploying a deep learning model for real-time inference on edge devices with limited power and memory. Which model optimisation technique would be MOST effective for reducing latency and memory footprint while maintaining acceptable accuracy?

Accepted Answer

Apply quantisation to convert weights from FP32 to INT8. Quantization reduces the precision of model weights from 32-bit floating point (FP32) to 8-bit integer (INT8), which directly cuts memory usage by 75% and accelerates inference on edge devices by leveraging integer arithmetic. This technique is specifically designed for resource-constrained environments where power and memory are limited, and it typically preserves accuracy within 1-2% of the original model.

Answer

Use a larger batch size during inference

Answer

Train the model for more epochs to improve convergence

Answer

Increase the number of layers to improve feature extraction

Question 13

A developer is using a large language model via an API. They want the model to solve a math problem step by step. Which prompt engineering technique should they use?

Accepted Answer

Chain-of-thought prompting. Chain-of-thought (CoT) prompting explicitly instructs the model to reason through intermediate steps before producing a final answer, which is exactly what's needed for multi-step math problems. By asking the model to 'think step by step,' it decomposes the problem into manageable reasoning stages, significantly improving accuracy on arithmetic and logic tasks.

Answer

Set temperature to 0.9

Answer

Few-shot prompting

Answer

Zero-shot prompting

Question 14

A security engineer is hardening an LLM-based API against OWASP LLM Top 10 risks. Which THREE risks should the engineer prioritize for mitigation?

Accepted Answer

Insecure output handling. Insecure output handling (A) is a core OWASP LLM Top 10 risk because LLM output passed to downstream systems without validation or sanitization can lead to XSS, SSRF, or remote code execution, so it must be prioritized. Training data poisoning (B) is also on the OWASP LLM Top 10 list, as manipulated or unvetted training/fine-tuning data can embed backdoors, bias, or malicious behavior into the model, directly threatening API integrity. Prompt injection (C) is the top-ranked OWASP LLM risk, since attackers can override system instructions or exfiltrate data through crafted inputs, making it essential for an LLM API hardening effort. Insecure deserialization (D) belongs to the OWASP Top 10 for web applications, not the LLM Top 10, and model quantization errors (E) are a model-performance/implementation concern rather than a recognized OWASP LLM Top 10 risk.

Answer

Insecure deserialization

Answer

Model quantization errors

Question 15

An organisation is developing a document intelligence system that extracts information from scanned invoices. Which THREE data preparation steps are critical to ensure high extraction accuracy? (Choose THREE.)

Accepted Answer

Cleaning and correcting OCR output. Option A (Cleaning and correcting OCR output) is correct because OCR on scanned invoices inevitably introduces character-level errors, and fixing those errors before feeding text to the extraction model directly improves field-level accuracy. Option D (Annotating bounding boxes and field labels) is correct because supervised document intelligence models need labelled ground truth that ties specific fields (e.g., invoice number, total) to their spatial locations to learn accurate extraction. Option E (Image preprocessing such as deskewing and binarisation) is correct because scanned invoices often suffer from rotation, noise, and uneven lighting, and correcting these at the image level raises OCR quality and downstream extraction accuracy. Option B (Removing punctuations and stopwords) is not appropriate because invoice fields such as dates, currency amounts, and vendor names rely on punctuation and specific tokens, so removing them would destroy critical information. Option C (Normalising all text to lowercase) is not appropriate because case can carry meaning in invoice data (e.g., currency codes, product identifiers, proper names), and lowercasing everything can reduce extraction fidelity.

Answer

Removing punctuations and stopwords

Answer

Normalising all text to lowercase