• Home
• Amazon Exams
• AWS Certified AI Practitioner AIF-C01

AWS Certified AI Practitioner AIF-C01 Amazon Practice Test Questions and Exam Dumps

Question No 1:

A manufacturing company generates quarterly forecasts using machine learning (ML) models to optimize its operations and meet projected customer demand.
An AI specialist is preparing a report for company stakeholders to ensure that the workings of the ML models are transparent and that the reasoning behind model predictions can be easily understood.

Which element should the AI specialist include in the report to best support transparency and explainability requirements?

A. Provide the full code used to train the machine learning models
B. Include Partial Dependence Plots (PDPs) for model interpretability
C. Share sample datasets that were used during model training
D. Present model convergence tables showing the training process

Answer: B. Partial Dependence Plots (PDPs)

Explanation:

When communicating machine learning results to non-technical stakeholders, transparency and explainability are crucial. Stakeholders need to understand why the model makes certain predictions, not just see the technical details behind the training process.

Partial Dependence Plots (PDPs) are powerful visualization tools that illustrate the relationship between one or two features and the predicted outcome. They show how changes in a specific input feature impact the model's predictions while holding other features constant. This greatly enhances a model’s interpretability, even for audiences without deep technical backgrounds.

• PDPs highlight which factors (such as sales numbers, seasonality, or operational variables) significantly influence forecasts.

• They allow stakeholders to visually grasp how the model behaves, making the decision-making process more trustworthy.
  
  

In contrast:

• A. Model training code: This is highly technical and unsuitable for a general business audience focused on understanding results.

• C. Sample training data: Provides context but doesn't explain how input features affect predictions.

• D. Convergence tables: Show optimization progress (e.g., loss decreasing) but do not offer insights into feature influence.

Thus, including Partial Dependence Plots in the report is the most effective way to meet the company’s transparency and explainability requirements, ensuring that stakeholders clearly understand how key features drive forecasting decisions.

Final Answer: B. Partial Dependence Plots (PDPs)

Question No 2:

A prestigious law firm plans to develop an AI-powered application leveraging Large Language Models (LLMs).
The goal of the application is to read complex legal documents and automatically extract the critical points and key information for easier review by legal teams.

Which AI solution should the law firm implement to meet these requirements?

A. Build an automatic Named Entity Recognition (NER) system
B. Create a recommendation engine to suggest legal actions
C. Develop a summarization chatbot that condenses document content
D. Develop a multi-language translation system for legal documents

Answer:  C. Develop a summarization chatbot

Explanation:

The primary requirement in this scenario is to extract key points and summarize legal documents for faster and more efficient review.
To achieve this, the most appropriate solution is to develop a summarization chatbot powered by Large Language Models (LLMs).

Summarization chatbots are designed to:

• Read and understand large volumes of text.

• Condense documents by identifying the main ideas, critical clauses, and essential facts.

• Deliver concise and coherent summaries without omitting crucial information.

LLMs like GPT or similar advanced models are well-suited for this task because they have been trained on massive text datasets and can understand, rephrase, and condense complex content — such as legal language — while maintaining accuracy.

Other options explained:

• A. Named Entity Recognition (NER): Useful for identifying specific entities (like names, places, dates) but does not summarize or extract key points.

• B. Recommendation engine: Typically suggests items based on user preferences, not suitable for document summarization.

• D. Translation system: Useful for language translation, not for extracting or summarizing key content from documents written in the same language.

Thus, building a summarization chatbot ensures that the firm’s legal teams can quickly grasp the essence of lengthy documents, saving significant time and enhancing decision-making processes.

Final Answer: C. Develop a summarization chatbot

Question No 3:

A biotechnology company is developing a machine learning (ML) system to classify human genes into 20 distinct categories based on their specific genetic characteristics.
A critical requirement is that the chosen ML model must provide clear transparency regarding how it makes its predictions — meaning the internal decision-making process must be understandable and explainable to researchers and stakeholders.

Which machine learning algorithm would best meet the company's needs for both classification and interpretability?

A. Decision Trees
B. Linear Regression
C. Logistic Regression
D. Neural Networks

Answer: A. Decision Trees

Explanation:

In this scenario, the company's priority is not just accurate classification but also explainability — meaning they need a model where the decision path for each prediction can be easily understood and traced.

Decision Trees are ideal for this purpose. They work by recursively splitting the data based on specific features (e.g., gene characteristics), resulting in a tree-like structure where each branch represents a decision rule. The path from the root to a leaf node explains exactly how a prediction was made.

Key benefits of using Decision Trees:

• High interpretability: Every decision at each node is human-readable.

• Transparent logic: Stakeholders can follow the decision process from input to output.

• Suitability for classification: They can handle multi-class classification problems like categorizing into 20 groups.

Why the other options are less suitable:

• Linear Regression: Used for predicting continuous values, not for classification.

• Logistic Regression: Good for binary classification and moderately interpretable, but handling 20 classes can become complex and less transparent.

• Neural Networks: Powerful for complex problems, but often referred to as "black boxes" due to their low interpretability — not ideal for situations where the internal mechanism must be easily documented.

Thus, to meet the company’s needs for clear explanation and multi-category classification, Decision Trees are the best choice.

Final Answer: A. Decision Trees

Question No 4:

A company has developed an image classification model to detect plant diseases by analyzing photos of plant leaves. The company is interested in evaluating the performance of the model and specifically wants to know how many images the model classified correctly.

Which evaluation metric should the company use to measure the model's accuracy?

A. R-squared score
B. Accuracy
C. Root Mean Squared Error (RMSE)
D. Learning rate

Answer:

B. Accuracy

Explanation:

In classification tasks, the goal is to assess how well the model can categorize input data into specific classes. Since the company is working with an image classification model designed to predict plant diseases from images, it needs an evaluation metric that will provide insight into how many images the model predicted correctly versus how many it got wrong.

The most appropriate evaluation metric for this scenario is Accuracy. Accuracy is defined as the percentage of correct predictions made by the model out of the total number of predictions. It is calculated as:

Accuracy=Number of correct predictionsTotal number of predictions\text{Accuracy} = \frac{\text{Number of correct predictions}}{\text{Total number of predictions}}Accuracy=Total number of predictionsNumber of correct predictions​

For example, if the model correctly classifies 90 out of 100 images, the accuracy would be 90%.

Why the other options are not ideal:

• R-squared score (Option A): Typically used for regression models, this metric assesses the proportion of variance in the target variable that is explained by the model. It is not suitable for classification tasks.

• Root Mean Squared Error (RMSE) (Option C): RMSE measures the average magnitude of error between predicted and actual values, typically used in regression tasks where the output is continuous. It's not appropriate for a classification problem.

• Learning Rate (Option D): The learning rate is a hyperparameter used to control the speed of training in optimization algorithms. It is not an evaluation metric for model performance.

Thus, for the image classification task, Accuracy is the most straightforward and meaningful metric to assess the number of correctly classified images.

Final Answer: B. Accuracy

Question No 5:

A company is utilizing a pre-trained large language model (LLM) to develop a chatbot that offers product recommendations. The company requires the LLM's responses to be concise and written in a specific language. 

Which of the following solutions would best ensure the chatbot’s responses align with the company's requirements for brevity and language specification?

Options:

A. Adjust the prompt.
B. Choose an LLM of a different size.
C. Increase the temperature.
D. Increase the Top K value.

Correct Answer: A. Adjust the prompt.

Explanation:

When using a pre-trained large language model (LLM) for specific tasks such as building a chatbot, it is crucial to fine-tune the model's behavior to meet particular requirements. In this case, the company needs the chatbot to generate responses that are not only concise but also in a specific language. The most effective way to ensure the model’s output aligns with these criteria is to adjust the prompt provided to the LLM.

Why Adjusting the Prompt is the Best Solution:

The prompt is the textual input or instruction given to the LLM, and it acts as a directive for generating responses. By customizing the prompt to explicitly state the company’s needs, the LLM can be guided to produce responses that meet the desired brevity and language. For instance, the prompt could be structured as:

• "Provide a short product recommendation in [specific language]."

• "Respond briefly in [specific language] with the best product for [customer need]."

This approach ensures that the LLM knows to prioritize short responses and adhere to the specified language, achieving the company's goals without requiring changes to the underlying model or additional configurations.

Why the Other Options are Less Effective:

• B. Choose an LLM of a Different Size: While larger models may produce more detailed or complex responses, they do not guarantee shorter, language-specific responses. A smaller model might suffice if the prompt is tailored correctly, and model size is unlikely to directly solve the problem.

• C. Increase the Temperature: The temperature parameter influences how creative and diverse the model’s responses are. Increasing the temperature makes the LLM more unpredictable, which could lead to longer or more varied responses, not necessarily shorter ones. This does not align with the company’s goal of generating concise answers.

• D. Increase the Top K value: The Top K parameter affects the range of possible word choices during response generation. While this influences the creativity and diversity of responses, it does not directly control the length or language-specific nature of the output.

In conclusion, the most efficient and targeted solution is to adjust the prompt, ensuring that the LLM generates short responses in the desired language by providing clear, concise instructions.

Question No 6: 

A company uses Amazon SageMaker for its machine learning (ML) pipeline in a production environment. The company processes large input data sizes up to 1 GB and experiences processing times of up to 1 hour. The company requires near real-time latency for inference. 

Which SageMaker inference option would best meet these requirements?

Options:

A. Real-time inference
B. Serverless inference
C. Asynchronous inference
D. Batch transform

Answer:

A. Real-time inference

Explanation:

In machine learning workflows, particularly when working with Amazon SageMaker, choosing the right inference option is critical to meet performance and latency requirements. For a company needing near real-time latency with large input data sizes and processing times, the optimal solution would be real-time inference.

Why Real-time Inference is the Best Solution:

Real-time inference in SageMaker is designed for scenarios where low-latency, on-demand predictions are required. This option allows the model to process incoming data and generate predictions immediately upon request, making it well-suited for applications where near-instant feedback is essential. Real-time inference can handle large input data sizes, typically up to several gigabytes, and can return responses within milliseconds to a few seconds, depending on the model complexity and infrastructure.

Given the company’s requirements—input data sizes of up to 1 GB and processing times of up to 1 hour—real-time inference can be configured to handle these demands by selecting appropriate infrastructure (such as multi-instance or GPU-based instances) that ensures both high processing power and fast response times. Additionally, SageMaker can scale in real-time inference to accommodate increased workloads, providing flexibility and performance optimization.

Why the Other Options Are Less Suitable:

• B. Serverless Inference: While serverless inference automatically scales based on the workload, it is more appropriate for smaller-scale, bursty workloads. It can incur additional latency during scaling and is not specifically optimized for handling large data sizes and long processing times like the 1 GB input data and 1-hour processing requirements described. Serverless inference works best for lightweight applications, making it unsuitable for this case.
  
  

• C. Asynchronous Inference: Asynchronous inference is designed for situations where predictions are not required immediately and can tolerate delays. It processes data in the background and provides results after some time, which does not align with the company’s requirement for near real-time latency.
  
  

• D. Batch Transform: This option is used for processing large datasets in batches, making it ideal for situations where predictions do not need to be delivered in real-time. Batch transform is better suited for scenarios where the input data size is large, but it does not provide the low-latency responses required by the company in a production environment.
  
  

In conclusion, real-time inference offers the best solution for applications that need to deliver low-latency predictions for large data sizes in a production setting.

Question No 7: 

A company is using domain-specific models for its machine learning tasks. Instead of creating new models from scratch, the company wants to adapt pre-trained models for related tasks. 

Which machine learning strategy would best meet these requirements?

A. Increase the number of epochs
B. Use transfer learning
C. Decrease the number of epochs
D. Use unsupervised learning

Answer:

B. Use transfer learning

Explanation:

The process of adapting pre-trained models to new, related tasks is known as transfer learning, and it is one of the most effective techniques for leveraging existing models to handle new challenges without starting from scratch.

Why Transfer Learning is the Best Strategy:

Transfer learning allows the company to use a model that has already been trained on a large dataset and fine-tune it for a new task, saving significant time and computational resources. This strategy is particularly beneficial when working with domain-specific models, as it enables the company to retain the learned features and knowledge from the original task while adapting the model to handle the nuances of the new task. It is often used when there is a shortage of labeled data for the new task, as the pre-trained model already understands many underlying patterns that can be applied to related tasks.

For example, a model pre-trained for natural language processing (NLP) tasks like sentiment analysis can be adapted through transfer learning to perform a different, but related task, such as text classification or named entity recognition. The company can freeze most of the initial layers and fine-tune the later layers of the network to specialize in the new task, which drastically reduces training time.

Why the Other Options Are Less Suitable:

• A. Increase the number of epochs: Increasing the number of epochs during training refers to allowing the model to iterate through the dataset more times, which can improve model performance in certain cases. However, this does not directly address the need for adapting pre-trained models. More epochs are useful for training from scratch but are less effective for fine-tuning a model that already has learned useful features.

• C. Decrease the number of epochs: Decreasing the number of epochs might result in underfitting, where the model does not have enough time to learn the new task properly. While this may reduce training time, it’s not a strategy for adapting pre-trained models to new tasks.

• D. Use unsupervised learning: Unsupervised learning involves training models on unlabeled data to discover patterns. While this is useful for tasks like clustering and anomaly detection, it does not directly help in adapting a pre-trained model to a new, related task. Transfer learning is specifically designed to solve this type of problem, making unsupervised learning unsuitable in this context.

In conclusion, transfer learning is the optimal strategy for adapting pre-trained models to new, related tasks, enabling the company to leverage existing knowledge while customizing models for specific needs.

Question No 8: 

A company is building a solution to generate images for protective eyewear. The solution must achieve high accuracy and minimize the risk of incorrect annotations. 

Which of the following solutions will best meet these requirements?

A. Human-in-the-loop validation by using Amazon SageMaker Ground Truth Plus
B. Data augmentation by using an Amazon Bedrock knowledge base
C. Image recognition by using Amazon Rekognition
D. Data summarization by using Amazon QuickSight Q

Answer:

A. Human-in-the-loop validation by using Amazon SageMaker Ground Truth Plus

Explanation:

To generate high-quality images for protective eyewear and minimize errors in annotations, the company needs a solution that combines automation and human oversight. Human-in-the-loop validation using Amazon SageMaker Ground Truth Plus is the best option for this task.

Why Human-in-the-loop Validation is the Best Choice:

Amazon SageMaker Ground Truth Plus offers a human-in-the-loop approach for machine learning tasks. This service allows automated labeling of images, with a human reviewer involved to validate and correct annotations. This setup ensures that the model’s predictions are accurate and that errors in the generated annotations are minimized. For tasks like image generation, where precision is critical, the combination of machine automation and human review guarantees a higher level of accuracy.

This solution is particularly useful for applications like the one in the question, where the company must ensure that generated images of protective eyewear are correctly labeled and do not contain errors that could impact subsequent model training or application deployment.

Why the Other Options Are Less Effective:

• B. Data augmentation by using an Amazon Bedrock knowledge base: While data augmentation can help expand the training dataset, it does not directly address the need for high accuracy or minimize annotation errors. Data augmentation methods alter existing data (like flipping or rotating images) to increase dataset variety but do not guarantee correct annotations.

• C. Image recognition by using Amazon Rekognition: Amazon Rekognition is excellent for image and video analysis but does not focus on generating images or minimizing annotation errors for tasks like the one described. It’s more suitable for analyzing existing images rather than generating them.

• D. Data summarization by using Amazon QuickSight Q: Amazon QuickSight Q is used for data visualization and querying, not image generation or annotation. It is not relevant to the task of creating accurate images and annotations for protective eyewear.

In conclusion, Amazon SageMaker Ground Truth Plus is the optimal solution because it combines automation with human oversight, ensuring both accuracy in image generation and minimized annotation errors.

Question No 9: 

A company is using a foundation model (FM) on Amazon Bedrock to create a chatbot. The FM needs to access encrypted data stored in an Amazon S3 bucket. The data is encrypted using Amazon S3 managed keys (SSE-S3). The FM encounters a failure when attempting to access the S3 bucket data. 

Which solution will resolve this issue?

A. Ensure that the role that Amazon Bedrock assumes has permission to decrypt data with the correct encryption key.
B. Set the access permissions for the S3 buckets to allow public access to enable access over the internet.
C. Use prompt engineering techniques to tell the model to look for information in Amazon S3.
D. Ensure that the S3 data does not contain sensitive information.

Answer:

A. Ensure that the role that Amazon Bedrock assumes has permission to decrypt data with the correct encryption key.

Explanation:

In this scenario, the failure encountered by the foundation model (FM) when trying to access encrypted data stored in an Amazon S3 bucket is due to insufficient permissions to decrypt the data. The model must have the correct permissions to access and decrypt the data encrypted with Amazon S3 managed keys (SSE-S3).

Why Option A is the Best Solution:

Amazon S3 uses encryption keys to protect data stored in buckets, and when you use SSE-S3, Amazon S3 manages the encryption and decryption process automatically. However, the IAM role assumed by Amazon Bedrock must have the necessary permissions to decrypt the data. This permission is granted by the correct IAM policy, which should include a statement that allows access to decrypt data using the SSE-S3 encryption key.

By ensuring that the role assumed by Amazon Bedrock has proper IAM permissions to decrypt the data, the FM will be able to access and process the encrypted data successfully.

Why the Other Options Are Less Effective:

• B. Set the access permissions for the S3 buckets to allow public access: Allowing public access to the S3 bucket is not recommended due to security concerns, especially when the data is sensitive. Public access can lead to unintended exposure of the data to unauthorized users.

• C. Use prompt engineering techniques to tell the model to look for information in Amazon S3: Prompt engineering does not address access control or encryption issues. It’s a technique used to guide the model’s behavior but will not resolve the issue of access permissions for encrypted data.

• D. Ensure that the S3 data does not contain sensitive information: This does not address the encryption issue or the failure to access the data. The problem lies in the permissions required for decryption, not the sensitivity of the data.

In conclusion, ensuring that the role assumed by Amazon Bedrock has the correct decryption permissions is the most effective solution to allow access to encrypted data stored in an S3 bucket.

Question No 10: 

A company wants to use language models for creating an application that performs inference on edge devices. The inference must have the lowest latency possible. 

Which solution will best meet these requirements?

Options:

A. Deploy optimized small language models (SLMs) on edge devices.
B. Deploy optimized large language models (LLMs) on edge devices.
C. Incorporate a centralized small language model (SLM) API for asynchronous communication with edge devices.
D. Incorporate a centralized large language model (LLM) API for asynchronous communication with edge devices.

Answer:

A. Deploy optimized small language models (SLMs) on edge devices.

Explanation:

When aiming for low-latency inference on edge devices, the primary goal is to ensure that the model can perform predictions quickly and efficiently without relying on external servers or APIs. This means choosing a solution that is optimized for fast, localized execution on the edge device itself.

Why Deploying Optimized Small Language Models (SLMs) is the Best Solution:

Small Language Models (SLMs) are specifically designed for low-latency, resource-constrained environments like edge devices. These models have fewer parameters compared to larger models, allowing them to run faster and consume less computational power. By deploying optimized small language models directly on the edge devices, inference can be performed locally, minimizing the delay typically associated with cloud-based or server-dependent systems.

Optimized small models are tailored for devices with limited memory, CPU, and battery power, ensuring that the language model can function efficiently even on devices with low resources, all while maintaining low latency for real-time inference.

Why the Other Options Are Less Effective:

• B. Deploy optimized large language models (LLMs) on edge devices: Large language models (LLMs) tend to require more computational resources (memory and processing power) than small models. Deploying these models on edge devices can introduce significant latency, as they demand more time to process and produce predictions.

• C. Incorporate a centralized small language model (SLM) API: While this option involves using a small model, it introduces latency due to the need for communication with a centralized server. This asynchronous model would not meet the need for low-latency inference as the edge devices would rely on cloud communication.

• D. Incorporate a centralized large language model (LLM) API: Like option C, using a large model in a centralized API introduces significant latency, as it involves round-trip communication to and from the cloud, which is slower than local inference on edge devices.

In conclusion, deploying optimized small language models (SLMs) on edge devices ensures the lowest possible latency for inference, making it the best solution for this scenario.