Free AIF-C01 Practice Test Questions and Answers (2026)

Benchmark datasets are specifically created and curated by the research community for evaluating model performance on specific tasks, including fairness and bias detection. These datasets are pre-labeled, cleaned, and structured, allowing for standardized and repeatable evaluations. Using an existing, relevant benchmark dataset (e.g., Civil Comments, Jigsaw Toxicity Datasets) eliminates the need for data collection, annotation, and structuring, which are time-consuming and resource-intensive tasks. This makes it the most efficient option with the least administrative overhead for the company to systematically assess the LLM's outputs for bias.

1. Dinan

E.

et al. (2020). Multi-dimensional Gender Bias Classification. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). This paper introduces a benchmark dataset specifically for evaluating gender bias

illustrating the role of such datasets. (See Section 3: "A New Dataset for Gender Bias Classification"

pp. 2-4). DOI: https://doi.org/10.18653/v1/2020.emnlp-main.391

2. Mehrabi

N.

et al. (2021). A Survey on Bias and Fairness in Machine Learning. ACM Computing Surveys

54(6)

1-35. This survey discusses evaluation methodologies

highlighting the use of benchmark datasets as a primary tool for auditing and quantifying bias in models. (See Section 4: "BIAS MITIGATION"). DOI: https://doi.org/10.1145/3457607

3. Stanford University. (2023). CS224N: Natural Language Processing with Deep Learning. Course materials frequently emphasize the use of standardized benchmark datasets (e.g.

GLUE

SQuAD) for model evaluation to ensure comparability and reproducibility

a principle that extends to fairness and bias evaluation. (See Lecture on "Model Evaluation").

Amazon Bedrock interacts with other AWS services, such as Amazon S3, by assuming an AWS Identity and Access Management (IAM) role. This role must be granted the specific permissions required to access the resources. When data in an S3 bucket is encrypted with SSE-S3, Amazon S3 manages the encryption and decryption keys. For a service like Bedrock to access an encrypted object, the IAM role it assumes must have s3:GetObject permission for that object. The failure indicates that the role lacks this necessary permission. Granting the permission allows Amazon S3 to transparently decrypt the data for Bedrock upon request, resolving the access failure.

1. AWS S3 Documentation on SSE-S3: "To access objects that are encrypted by using SSE-S3

you must have s3:GetObject permission. When you have this permission

Amazon S3 handles both encryption and decryption for you. You don't need to make any changes to your applications to access SSE-S3 encrypted objects."

Source: AWS Security Documentation

Amazon Simple Storage Service User Guide

"Protecting data using server-side encryption with Amazon S3-managed encryption keys (SSE-S3)"

Section: "Accessing objects that are encrypted using SSE-S3".

2. AWS Bedrock Documentation on Data Source Permissions: "For Amazon S3 data sources

Amazon Bedrock requires permissions to access the S3 bucket containing your files. You can create a service role that gives Amazon Bedrock these permissions... The role must have a trust policy that allows Amazon Bedrock to assume it... Attach an identity policy to the role that grants permissions to your S3 bucket." The documentation then specifies that the policy must include the s3:GetObject action.

Source: AWS Foundation Model Service Documentation

Amazon Bedrock User Guide

"Knowledge bases for Amazon Bedrock"

Section: "Create a data source".

3. AWS IAM Documentation on Service Roles: "An IAM role is an IAM identity that you can create in your account that has specific permissions... A role can be assumed by anyone who needs it... An AWS service can assume a role to perform actions in other services on your behalf."

Source: AWS Identity and Access Management Documentation

IAM User Guide

"IAM roles"

Section: "IAM role use cases".

The described activities—creating a correlation matrix, calculating summary statistics, and visualizing data—are the core components of Exploratory Data Analysis (EDA). The primary goal of EDA is to understand the main characteristics of a dataset, uncover patterns, identify anomalies, and check assumptions before any modeling is performed. This stage is crucial for gaining initial insights that will guide subsequent steps like feature engineering and data pre-processing.

1. AWS Whitepaper

"The AWS Machine Learning Lens": This document outlines the machine learning lifecycle. In the section "Phase 1: Business goal and problem formulation

" it describes the data exploration and analysis step: "You perform exploratory data analysis (EDA) to understand the data and extract insights. This includes data profiling

visualization

and statistical analysis." (AWS Machine Learning Lens

Feb 2023

Page 10).

2. University Courseware

Stanford University CS229: The course materials for Machine Learning (CS229) consistently position EDA as a foundational step. Lecture notes describe this phase as the process of "looking at your data" using visualizations and summary statistics to understand its properties before attempting to train a model. (Ng

A. (2023). CS229 Machine Learning Course Materials

Stanford University

Lecture Notes 1

Section on "The Machine Learning Workflow").

3. Academic Publication

"CRISP-DM 1.0: Step-by-step data mining guide": The Cross-Industry Standard Process for Data Mining (CRISP-DM) is a widely accepted methodology. Its second phase

"Data Understanding

" directly corresponds to EDA. This phase includes tasks such as "Compute basic statistics" and "Visualize data" to gain familiarity with the data and identify quality problems. (Chapman

P.

et al. (2000). CRISP-DM 1.0: Step-by-step data mining guide

The CRISP-DM Consortium

Section 3.2 "Data Understanding Phase").

The core task is to segment customers into tiers using a dataset that is explicitly described as "unlabeled." Unsupervised learning is the machine learning paradigm designed specifically for this type of problem. It aims to find hidden patterns, structures, or groupings within data that has no predefined labels or target outcomes. The process of classifying customers into tiers based on their inherent similarities is a classic example of clustering, which is a primary technique within unsupervised learning. This approach allows the company to discover natural customer segments directly from the data itself.

1. Ng

A. (2023). CS229 Machine Learning Course Notes. Stanford University. In Part V

"Unsupervised Learning

" it is stated: "In the unsupervised learning problem

we are given a dataset and we are told to find some structure in the data. A canonical example of unsupervised learning is clustering

in which we are given a set of unlabeled data and we want to group the data into 'cohesive' clusters." (p. 1).

2. Amazon Web Services. (n.d.). What is Unsupervised Learning?. AWS Machine Learning Documentation. In the section "Unsupervised learning use cases

" it specifies: "A common application of unsupervised learning is customer segmentation. You can use it to group customers based on their purchasing behavior

demographics

or other characteristics... to tailor marketing campaigns."

3. Hastie

T.

Tibshirani

R.

& Friedman

J. (2009). The Elements of Statistical Learning: Data Mining

Inference

and Prediction (2nd ed.). Springer. In Chapter 14

"Unsupervised Learning

" the text introduces the topic: "In the unsupervised learning problem we are given a set of observations... but are not told anything about them... The goal is to discover interesting things about the measurements... to group them into clusters." (Section 14.3

p. 485).

Prompt chaining is a technique used to manage complex tasks by decomposing them into a series of smaller, manageable subtasks. Each subtask is handled by a distinct prompt sent to the large language model (LLM). The output of one prompt in the sequence is then used as part of the input for the next prompt. This sequential, step-by-step process allows the LLM to build on intermediate results, enabling it to handle intricate workflows and multi-step reasoning that would be too complex to address in a single, monolithic prompt.

1. Google Cloud. (2024). Introduction to prompt design. Generative AI Documentation. In the "Prompt techniques" section

under the subsection "Chaining

" it states: "Chaining is the technique of breaking a single

large task into a series of smaller

interconnected sub-tasks... The output of one prompt is used in the input of a subsequent prompt."

2. Microsoft. (2024). Introduction to prompt engineering. Azure AI services documentation. In the "Prompt engineering techniques" section

under the subsection "Chaining

" it describes the technique as: "Chaining involves breaking down a large task into a series of smaller

connected prompts. The output of one prompt becomes the input for the next."

3. Wei

J.

Wang

X.

Schuurmans

D.

Bosma

M.

Chi

E.

Le

Q.

& Zhou

D. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems

35

24824-24837. The paper's core concept (Section 2

"Chain-of-Thought Prompting") is a specific implementation of sequential prompting where intermediate reasoning steps are generated

demonstrating the principle of breaking down a problem sequentially.

4. Zhao

W. X.

Zhou

K.

Li

J.

Tang

T.

Wang

X.

Hou

Y.

... & Wen

J. R. (2023). A Survey of Large Language Models. arXiv preprint arXiv:2303.18223. Section 4.2.2

"Chain-of-Thought (CoT) Prompting

" describes this method as generating "a sequence of intermediate reasoning steps

" which is a foundational concept for prompt chaining. (DOI: https://doi.org/10.48550/arXiv.2303.18223)

In artificial intelligence, an embedding is a learned representation for discrete or categorical data, such as words, images, or user IDs. It maps these high-dimensional, often sparse inputs into a dense, continuous-valued vector in a lower-dimensional space. The key characteristic of this vector space is that it captures semantic relationships; items with similar meanings or functions are positioned closer to one another. This numerical representation is more efficient and meaningful for machine learning models to process compared to raw, high-dimensional data like one-hot encoded vectors.

1. University Courseware (Stanford): Jurafsky

D.

& Martin

J. H. (2023). Speech and Language Processing (3rd ed. draft). In Chapter 6

"Vector Semantics and Embeddings

" the text defines embeddings as "a short

dense vector for each word

" contrasting them with sparse

high-dimensional vectors. The chapter focuses on how these vectors represent meaning in a geometric space. (See Chapter 6

Section 6.3).

2. Official Vendor Documentation (TensorFlow): Google. (n.d.). Embeddings. TensorFlow Core. Retrieved from tensorflow.org. The official documentation states

"An embedding is a mapping from discrete objects

such as words

to vectors of real numbers... The embedding is a lookup table that maps from integer indices (which stand for specific words) to dense vectors (their embeddings)." (See the "Introduction" section).

3. Academic Publication: Mikolov

T.

Chen

K.

Corrado

G.

& Dean

J. (2013). Efficient Estimation of Word Representations in Vector Space. arXiv:1301.3781. The foundational Word2Vec paper introduces the concept of learning "high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships." This describes the core purpose of embeddings. (See Abstract and Section 1: Introduction).

4. University Courseware (MIT): MIT 6.S191: Introduction to Deep Learning. (2021). Lecture 4: Sequence Modeling. The lecture describes embeddings as a way to "learn a low-dimensional

dense representation of words" to overcome the sparsity and lack of relationship information in one-hot vectors. (Available on YouTube and course website

discussion typically occurs in the first 15-20 minutes of the NLP/Sequence lecture).

AWS Data Exchange is a service designed to facilitate the exchange of data between data providers and subscribers. In this scenario, the Independent Software Vendors (ISVs) can act as data providers, publishing their compliance reports as data products. The AI company can subscribe to these products. When a new report is published, AWS Data Exchange generates an event in Amazon EventBridge. This event can be used to trigger an AWS Lambda function or an Amazon Simple Notification Service (SNS) topic to send an email notification to the company, directly meeting the specified requirement.

1. AWS Data Exchange Documentation: "Subscribing to data products in AWS Data Exchange." This page describes the core functionality of subscribing to third-party data. The section "Tutorial: Subscribing to a product" outlines the process a customer would follow.

Source: AWS Data Exchange User Guide.

2. AWS Data Exchange Documentation: "Monitoring AWS Data Exchange events with Amazon EventBridge." This document explicitly states

"When a provider publishes a new revision to a data set that you're subscribed to

AWS Data Exchange sends an event to Amazon EventBridge." It further explains how to create rules to react to these events

such as sending notifications.

Source: AWS Data Exchange User Guide.

3. AWS Audit Manager Documentation: "What Is AWS Audit Manager?" The first paragraph states

"AWS Audit Manager helps you continuously audit your AWS usage to simplify how you assess risk and compliance with regulations and industry standards." This confirms its purpose is for auditing one's own environment.

Source: AWS Audit Manager User Guide.

4. AWS Artifact Documentation: "What is AWS Artifact?" The introduction clarifies

"AWS Artifact is your go-to

central resource for compliance-related information... It provides on-demand access to AWS’ security and compliance reports..." This specifies the source of reports as AWS itself.

Source: AWS Artifact User Guide.

The scenario describes an application that must understand and process human language—specifically, written customer feedback—to classify it into predefined categories. This task of enabling computers to process, analyze, and derive meaning from textual data is the core function of Natural Language Processing (NLP). Text classification, as described in the question, is a fundamental and widely used NLP technique.

1. Jurafsky

D.

& Martin

J. H. (2023). Speech and Language Processing (3rd ed. draft). In Chapter 4

"Naive Bayes

Text Classification

and Sentiment

" the authors introduce text classification as a primary task of NLP: "The task of text classification is to take a text... and assign it a class... from a fixed set of classes." The scenario of classifying feedback into categories like "product quality" is a direct example of this.

2. Amazon Web Services (AWS). (n.d.). What is Natural Language Processing (NLP)? AWS documentation defines NLP as a field of AI that "focuses on enabling computers to understand

interpret

and manipulate human language." It lists text classification and sentiment analysis (a related task) as key applications.

3. Manning

C. D. (2011). CS224N: Natural Language Processing with Deep Learning

Lecture 1: Introduction and Word Vectors. Stanford University. The introductory lecture defines NLP as the field that gives "machines the ability to read

understand and derive meaning from human languages." The task of classifying customer feedback fits squarely within this definition.

4. Aggarwal

C. C.

& Zhai

C. (2012). A Survey of Text Classification Algorithms. In C. C. Aggarwal & C. Zhai (Eds.)

Mining Text Data (pp. 163-222). Springer. Section 1.1

"Introduction

" describes text classification as "the problem of assigning a document to one or more predefined categories

" which directly aligns with the scenario presented in the question. DOI: https://doi.org/10.1007/978-1-4614-3223-46

Amazon Textract is a machine learning service that automatically extracts text, handwriting, and data from scanned documents, including PDFs. It goes beyond simple Optical Character Recognition (OCR) to identify the structure of the document, such as forms and tables. This service directly addresses the company's need to convert PDF resumes into a plain text format for further processing, automating a previously manual task.

1. Amazon Textract Developer Guide: "Amazon Textract is a machine learning (ML) service that automatically extracts text

handwriting

and data from scanned documents. It goes beyond simple optical character recognition (OCR) to identify

understand

and extract data from forms and tables."

Source: AWS Documentation

What Is Amazon Textract?

Amazon Textract Developer Guide.

2. Amazon Personalize Developer Guide: "Amazon Personalize is a machine learning service that makes it easy for developers to create individualized recommendations for customers using their applications."

Source: AWS Documentation

What is Amazon Personalize?

Amazon Personalize Developer Guide.

3. Amazon Lex V2 Developer Guide: "Amazon Lex is an AWS service for building conversational interfaces for applications using voice and text."

Source: AWS Documentation

What is Amazon Lex V2?

Amazon Lex V2 Developer Guide.

4. Amazon Transcribe Developer Guide: "Amazon Transcribe is an automatic speech recognition (ASR) service that makes it easy for you to add speech-to-text capabilities to your applications."

Source: AWS Documentation

What is Amazon Transcribe?

Amazon Transcribe Developer Guide.

The model demonstrates high performance on the training data (90% recall) but significantly lower performance on the unseen testing data (40% recall). This large discrepancy is a classic symptom of overfitting. Overfitting occurs when a machine learning model learns the training data too well, including its noise and specific patterns, to the point where it negatively impacts the model's ability to generalize to new, unseen data. The model has effectively "memorized" the training set instead of learning the underlying, generalizable trends.

1. Amazon SageMaker Developer Guide: In the section on model tuning

it is stated that a sign of overfitting is when the model performs well on the training set but poorly on the validation set. "If the training accuracy is high and the validation accuracy is low

the model is likely overfitting to the training dataset."

Source: Amazon SageMaker Developer Guide

"Monitor and Analyze Training Jobs Using Metrics

" section on "Look for overfitting or underfitting".

2. Stanford University

CS229 Machine Learning Course Notes: The course materials define overfitting as a state where a model has a low training error but a high generalization (test) error. This discrepancy indicates the model has failed to capture the true underlying pattern and instead modeled the noise in the training data.

Source: Ng

A. (2023). CS229 Machine Learning Course Notes. Stanford University. Part V

"Learning Theory

" Section on Bias/variance tradeoff.

3. MIT OpenCourseWare

Introduction to Machine Learning: Lecture notes explain that overfitting happens when a model is too complex for the amount of training data available. This results in a model that fits the training data points almost perfectly but fails to generalize to new data points

leading to a large gap between training and test performance.

Source: MIT OpenCourseWare. (2020). 6.036 Introduction to Machine Learning

Fall 2020. Lecture 9 Notes: "Regularization

" Section 9.1.

Amazon SageMaker Clarify provides tools to gain deeper insights into machine learning models and data. One of its core functionalities is the ability to detect potential statistical bias in a dataset before model training begins. By computing various pre-training bias metrics (e.g., Class Imbalance, Difference in Proportions of Labels), data scientists can identify and mitigate fairness issues early in the ML lifecycle, during the data preparation phase. This helps ensure that the data used to train the model is balanced and fair across different subgroups.

1. Amazon SageMaker Developer Guide: Under the section "Fairness and Explainability with Amazon SageMaker Clarify

" the guide states

"With SageMaker Clarify

you have the option to measure pre-training bias on your dataset and post-training bias on your model." This directly supports the functionality of identifying bias during the data preparation (pre-training) stage.

2. Amazon SageMaker Developer Guide: The section "Run a SageMaker Clarify Job for Pre-training Bias Analysis" provides a detailed walkthrough

stating

"To check for biases in the initial dataset

you must configure and run a SageMaker Clarify processing job." This confirms option D as a primary use case.

3. Amazon SageMaker Developer Guide: The page "Monitor models for data and model quality" describes the role of SageMaker Model Monitor

clarifying the distinction from Clarify: "Amazon SageMaker Model Monitor automatically monitors machine learning (ML) models in production... It can monitor for drifts in data quality

model quality

bias

and feature attribution." This shows that general quality monitoring is the role of Model Monitor

making option B less precise.

4. Amazon SageMaker Developer Guide: The section "Amazon SageMaker Model Cards" explains its purpose: "Amazon SageMaker Model Cards help you document critical details about your machine learning (ML) models in a single place for streamlined governance and reporting." This confirms that option C describes Model Cards

not Clarify.

Embeddings are learned, dense numerical representations (vectors) of discrete data, such as words, sentences, or entire documents. In natural language processing (NLP), these vectors capture the semantic meaning and context of the textual information. The model places words with similar meanings close to each other in a multi-dimensional vector space. This allows AI models to understand relationships, context, and nuances in language, which is a significant improvement over sparse, high-dimensional representations like one-hot encoding. For example, the vectors for "king" and "queen" would be located in a similar region of the vector space.

1. Google Developers. (n.d.). Machine Learning Glossary: Embedding. "An embedding is a relatively low-dimensional space into which you can translate high-dimensional vectors. Embeddings make it easier to do machine learning on large inputs like sparse vectors representing words." Retrieved from Google's official developer documentation.

2. Mikolov

T.

Chen

K.

Corrado

G.

& Dean

J. (2013). Efficient Estimation of Word Representations in Vector Space. arXiv:1301.3781. In the introduction (Section 1)

the paper states

"we propose two new model architectures for computing continuous vector representations of words... The main contribution of this paper is a new technique that allows the training of high-quality word vectors from huge data sets." This foundational paper introduces the concept of learning these numerical representations.

3. Manning

C.

Jurafsky

D.

& Potts

C. (2023). CS224N: Natural Language Processing with Deep Learning. Stanford University. In Lecture 2: "Word Vectors and Word Senses

" it is explained that a word embedding w is a "dense vector for each word

chosen so that it is similar to vectors of words that appear in similar contexts." (Slide 11).

The most cost-effective and scalable solution is to use a Knowledge Base for Amazon Bedrock. This feature is specifically designed for Retrieval-Augmented Generation (RAG), a technique ideal for this scenario. It works by ingesting the PDF documents, converting them into a searchable vector format, and storing them. At query time, the system retrieves only the most relevant snippets of information from the manuals to augment the user's prompt. This keeps the input to the Large Language Model (LLM) concise and relevant, which significantly reduces the number of tokens processed per query and, therefore, minimizes inference costs. This method avoids the high costs associated with fine-tuning or sending large, irrelevant context in every prompt.

1. Official Vendor Documentation: AWS Documentation

"Knowledge bases for Amazon Bedrock." The documentation states

"Knowledge Bases for Amazon Bedrock is a fully managed capability that helps you implement the entire RAG workflow... without having to build custom integrations to data sources." This directly describes the solution in option D.

2. Official Vendor Documentation: AWS Machine Learning Blog

"Choose the right model customization method in Amazon Bedrock." In the section comparing RAG and fine-tuning

it notes

"RAG can be more cost-effective than fine-tuning

especially for models that are used infrequently... RAG is a good choice when your model needs to access rapidly changing information." This supports the cost-effectiveness of the RAG approach (implemented by Knowledge Bases) over fine-tuning (Option C).

3. Peer-Reviewed Academic Publication: Lewis

P.

et al. (2020). "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks." Advances in Neural Information Processing Systems 33

Section 1

Paragraph 2. This foundational paper introduces the RAG framework

which combines a pre-trained retriever with a pre-trained generator. The Knowledge Base in Bedrock is a managed implementation of this exact

efficient pattern. (Available via arXiv:2005.11401)

4. University Courseware: Stanford University

CS224N - NLP with Deep Learning

Winter 2023

Lecture 17

"Retrieval and Question Answering." The lecture materials discuss RAG as a state-of-the-art method for open-domain question answering

highlighting its ability to ground models in external knowledge without costly retraining

which aligns with the cost-effectiveness requirement.

Building a well-architected ML workload follows a structured lifecycle as outlined in AWS best practices. The process begins with defining the business goal and framing the ML problem to ensure the project aligns with organizational objectives. Next, the model is developed, which includes data preparation, training, and evaluation. Once the model is ready, it is deployed tomake predictions in a production environment. Finally, the model is monitored to ensure it performs as expected and to address any issues like drift or degradation over time. This order ensures a systematic approach to ML development. Exact Extract from AWS AI Documents: From the AWS AI Practitioner Learning Path: "The machine learning lifecycle typically follows these stages: 1) Define the business goal and frame the ML problem, 2) Develop the model (including data preparation, training, and evaluation), 3) Deploy the model to production, and 4) Monitor the model for performance and drift to ensure it continues to meet business needs." (Source: AWS AI Practitioner Learning Path, Module on Machine Learning Lifecycle) Detailed Step 1: Define business goal and frame ML problemThis is the first step in any ML project. It involves understanding the business objective (e.g., reducing churn) and framing the ML problem (e.g., classification or regression). Without this step, the project lacks direction. The hotspot lists this option as "Define business goal and frame ML problem," which matches this stage. Step 2: Develop modelAfter defining the problem, the next step is to develop the model. This includes collecting and preparing data, selecting an algorithm, training the model, and evaluating its performance. The hotspot lists "Develop model" as an option, aligning with this stage. Step 3: Deploy modelOnce the model is developed and meets performance requirements, it is deployed to a production environment to make predictions or automate decisions. The hotspot includes "Deploy model" as an option, which fits this stage. Step 4: Monitor modelAfter deployment, the model must be monitored to ensure it performs well over time, addressing issues like data drift or performance degradation. The hotspot lists "Monitor model" as an option, completing the lifecycle. Hotspot Selection Analysis: The hotspot provides four steps, each with the same dropdown options: "Select...," "Deploy model," "Develop model," "Monitor model," and "Define business goal and frame ML problem." The correct selections are: Step 1: Define business goal and frame ML problem Step 2: Develop model Step 3: Deploy model Step 4: Monitor model Each option is used exactly once, as required, and follows the logical order of the ML lifecycle. Reference: AWS AI Practitioner Learning Path: Module on Machine Learning Lifecycle Amazon SageMaker Developer Guide: Machine Learning Workflow (https://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works-mlconcepts.html) AWS Well-Architected Framework: Machine Learning Lens (https://docs.aws.amazon.com/wellarchitected/latest/machine-learning-lens/)

A Retrieval-Augmented Generation (RAG) pipeline is typically divided into two distinct phases: an offline data preparation/indexing phase and an online inference/querying phase. To ensure a near real-time user experience, operations that are computationally intensive and independent of the user's immediate query are performed offline in batch.

The generation of embeddings for the entire corpus of published content (A) and the subsequent creation or updating of the search index (C) are part of this offline phase. Since new content is added daily, these steps can be scheduled as a recurring batch job to process the new documents and update the vector store without impacting the performance of the live user-facing application. The remaining steps are part of the online inference pipeline that must execute quickly upon receiving a user query.

1. Lewis

P.

et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems (NeurIPS) 33.

Section 3

"METHODS": The paper describes the non-parametric memory

which is "a dense vector index of Wikipedia

built using the Dense Passage Retriever (DPR)". This "building" of the index from the content corpus is an offline

prerequisite step before any queries are processed.

2. Stanford University

CS224N: NLP with Deep Learning | Winter 2023.

Lecture 16: "Retrieval and Question Answering": The lecture on Dense Passage Retrieval (DPR)

a common retriever in RAG

explicitly distinguishes between the offline and online processes. It details the "offline pre-computation" step where all passages (content) are encoded into embeddings and indexed. The "at test time" (online) process involves encoding the question and performing a fast nearest neighbor search.

3. AWS Documentation

"Retrieval Augmented Generation (RAG) on AWS".

Architectural Overview: Official AWS architectural patterns for RAG consistently show a "Data Ingestion and Processing" pipeline that is separate from the "Inference" pipeline. The ingestion pipeline

which runs on a schedule (e.g.

daily)

is responsible for chunking documents

generating embeddings

and loading them into a vector store like Amazon OpenSearch Service or Amazon Kendra. This is explicitly described as a batch process.

The core requirement is to perform sentiment analysis on written text from customer reviews.

Amazon Comprehend is a managed Natural Language Processing (NLP) service specifically designed for this type of task. It has a dedicated API for sentiment analysis that can identify whether the sentiment of a text is positive, negative, neutral, or mixed.

Amazon Bedrock provides access to a range of powerful foundation models (FMs) from leading AI companies. These large language models (LLMs) are highly proficient at understanding context and nuance in text, making them well-suited for performing sophisticated sentiment analysis as part of a broader generative AI application.

1. Amazon Comprehend: According to the official documentation

"Amazon Comprehend uses natural language processing (NLP) to extract insights about the content of documents... The sentiment analysis operation determines the overall sentiment of a text (Positive

Negative

Neutral

or Mixed)."

Source: AWS Documentation

"Amazon Comprehend Developer Guide

" section: "Sentiment analysis."

2. Amazon Bedrock: The documentation states

"Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs)..." These FMs

such as Amazon Titan Text and Anthropic's Claude

are designed for a wide range of NLP tasks

including text classification and sentiment analysis.

Source: AWS Documentation

"Amazon Bedrock User Guide

" section: "What is Amazon Bedrock?" and "Foundation models."

3. Amazon Lex (Incorrect): The official guide states

"Amazon Lex is an AWS service for building conversational interfaces into any application using voice and text." This confirms its purpose is for interaction

not bulk text analysis.

Source: AWS Documentation

"Amazon Lex V2 Developer Guide

" section: "What Is Amazon Lex?".

To manage and secure the flow of data between Amazon SageMaker Studio and Amazon S3, the best practice is to configure SageMaker to operate within a Virtual Private Cloud (VPC). By creating a VPC endpoint for S3, you ensure that the data traffic between your SageMaker Studio notebooks and your S3 bucket does not traverse the public internet. Instead, the data flows over the private AWS network. This provides a secure, controlled, and often more performant data path, which directly addresses the requirement to manage the data flow. Endpoint policies can be attached to further restrict access to specific S3 buckets.

1. AWS SageMaker Developer Guide: "To control access to your data

we recommend that you create a private VPC and configure it so that your data is not accessible over the internet. To allow your Studio notebooks to access the Amazon S3 buckets where you store your data... create a VPC endpoint for Amazon S3."

Source: AWS SageMaker Developer Guide

Section: "Securing Amazon SageMaker Studio connectivity". Document path: SageMaker > SageMaker Studio > Administer Amazon SageMaker Studio > Securing Amazon SageMaker Studio connectivity.

2. AWS VPC User Guide: "A VPC endpoint enables you to privately connect your VPC to supported AWS services... without requiring an internet gateway

NAT device

VPN connection

or AWS Direct Connect connection. Instances in your VPC do not require public IP addresses to communicate with resources in the service. Traffic between your VPC and the other service does not leave the Amazon network."

Source: AWS Virtual Private Cloud User Guide

Section: "VPC endpoints".

3. AWS Whitepaper - Building a Secure Data Lake on AWS: This whitepaper discusses architectural patterns for securing data. It states

"Use VPC endpoints to access data in Amazon S3 from your VPC without using public IPs and without traffic traversing the internet." This principle applies directly to securing the connection from a SageMaker instance within a VPC.

Source: AWS Whitepaper: "Building a Secure Data Lake on AWS"

Page 11

Section: "Network and Host Security".

Amazon QuickSight is the AWS business intelligence (BI) service designed for creating interactive dashboards and data visualizations. The integration of Amazon Q within QuickSight specifically enables users to ask questions in natural language (e.g., "Show total sales for top products in the last 12 months") and have the service automatically generate the relevant graphs and visuals. This capability directly addresses the company's requirement to automate the generation of graphs from their sales data without needing to manually build reports.

1. Amazon QuickSight User Guide: "With Amazon Q in Amazon QuickSight

business users can ask questions about their data in natural language and receive accurate answers with relevant visualizations

helping them to understand their data better."

Source: AWS Documentation

Amazon QuickSight User Guide

"Building visuals with Amazon Q in Amazon QuickSight".

2. Amazon Q Developer Documentation: "Amazon Q Developer is a generative AI-powered assistant that can help you with a wide variety of tasks throughout your software development lifecycle (SDLC)."

Source: AWS Documentation

Amazon Q Developer User Guide

"What is Amazon Q Developer?".

3. Amazon Q in AWS Chatbot: "Amazon Q in AWS Chatbot helps you get answers to your questions about AWS services

architecture

and best practices... You can use Amazon Q to troubleshoot issues

get guidance on how to configure your AWS resources

and more."

Source: AWS Documentation

AWS Chatbot Administrator Guide

"Using Amazon Q in AWS Chatbot".

4. Carnegie Mellon University

Cloud Computing Courseware: Course materials on cloud-based data analytics often distinguish between developer tools and BI platforms. They categorize services like QuickSight as BI tools for end-user reporting and visualization

separate from developer-centric AI assistants.

Source: Carnegie Mellon University

15-319/619 Cloud Computing

Fall 2023

Lecture 19: Data Analytics & Big Data. (Illustrates the distinct purpose of BI services).

The foundation model (FM) is underperforming because it lacks familiarity with the specialized vocabulary ("complex scientific terms") of the research domain. Domain adaptation fine-tuning is the process of further training a pre-trained model on a specific, unlabeled dataset—in this case, the research papers. This process adjusts the model's internal parameters (weights) to learn the nuances, terminology, and statistical patterns of the scientific domain. This directly addresses the root cause of the problem by enhancing the model's core comprehension of the specialized content, leading to significantly improved performance.

1. AWS Documentation: The Amazon Bedrock User Guide explains that fine-tuning adapts a model for specific tasks or domains. It states

"Fine-tuning is the process of taking a pre-trained foundation model (FM) and further training it on your own dataset... to make it more specialized for your specific application."

Source: Amazon Bedrock User Guide

"Custom models

" section on "Fine-tuning."

2. University Courseware: Stanford University's course on Large Language Models distinguishes between in-context learning (prompting) and fine-tuning. Fine-tuning modifies the model's weights to specialize it

which is necessary when the task requires deep domain knowledge that cannot be conveyed in a few examples.

Source: Stanford University

CS324: Large Language Models

Winter 2022

Lecture 3: "Capabilities

" section on "Adaptation."

3. Academic Publication: A foundational paper on language models explains that fine-tuning is a critical step for adapting large pre-trained models to specific downstream tasks or domains

which significantly improves performance over using the base model alone.

Source: Devlin

J.

Chang

M. W.

Lee

K.

& Toutanova

K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies

Volume 1 (Long and Short Papers)

pp. 4171–4186. Section 4: "Experiments." (https://doi.org/10.18653/v1/N19-1423)

Amazon SageMaker JumpStart is a machine learning hub that accelerates the ML journey by providing access to hundreds of pre-trained models, including state-of-the-art foundation models (FMs). It allows development teams to select an FM and deploy it with a single click, which automatically creates a secure, scalable Amazon SageMaker endpoint. This endpoint can be configured to be accessible only from within the team's Virtual Private Cloud (VPC), satisfying the requirement for quick deployment and private consumption.

1. Amazon SageMaker Developer Guide: "Amazon SageMaker JumpStart helps you quickly and easily get started with machine learning. JumpStart provides one-click deployment and fine-tuning of a wide variety of pre-trained models from popular model hubs

including foundation models."

Source: AWS Documentation

Amazon SageMaker Developer Guide

"Amazon SageMaker JumpStart".

2. Amazon SageMaker Developer Guide: "To control access to your models

we recommend that you configure your SageMaker to use a private VPC... When you use a private VPC

you can configure it so that your model containers aren't accessible over the internet."

Source: AWS Documentation

Amazon SageMaker Developer Guide

"Protect Endpoints by Using a Virtual Private Cloud".

3. AWS Machine Learning Blog: "With SageMaker JumpStart

you can choose from a growing list of best-performing foundation models... With one-click deployment

you can get a dedicated endpoint for your chosen model and use it in your applications."

Source: AWS Machine Learning Blog

"Amazon SageMaker JumpStart simplifies and accelerates access to foundation models for generative AI"

May 25

2023.