Commonly Confused Services on AIF-C01

Deciding signal

Amazon Bedrock provides API access to pre-trained foundation models (FMs) from AWS and third-party providers — Anthropic Claude, Meta Llama, Amazon Titan, and others. You call an API; AWS manages the model infrastructure. It is the right answer when the scenario involves building applications on top of existing large language models without training or managing models. SageMaker is a platform for building, training, tuning, and deploying custom machine learning models. It provides compute infrastructure, managed notebooks, training jobs, hyperparameter tuning, and model hosting. It is the right answer when the scenario involves training a model on proprietary data, managing the ML training pipeline, or deploying a custom model. The decisive signal is whether a pre-trained foundation model is sufficient or whether custom model training is required.

Deciding signal

Rekognition analyzes images and videos — detecting objects, faces, text in images, celebrities, unsafe content, and custom labels. The input is always an image or video file. Comprehend is a natural language processing service that analyzes text: sentiment, entities, key phrases, language detection, and topic modeling. The input is text. Textract extracts structured data from documents — forms, tables, and handwriting — going beyond simple OCR to understand the document layout. The input is a document image (PDF, TIFF, JPEG). Transcribe converts speech in audio or video files to text. The input is audio. The input modality is the primary discriminator: video/image (Rekognition), audio (Transcribe), document layout (Textract), text content (Comprehend).

Deciding signal

Amazon Q Business is a generative AI assistant for employees — it connects to enterprise data sources (SharePoint, S3, Salesforce, Confluence) and answers questions about business content, policies, and documentation. It is a retrieval-augmented generation (RAG) application targeted at business end-users. Amazon Q Developer is an AI coding assistant integrated into IDEs and the AWS Console — it generates code, explains AWS services, reviews security vulnerabilities, and assists with infrastructure as code. The user is a developer; the use case is software development productivity. When the scenario involves employees asking questions about company policies or internal documents, Q Business. When it involves developers getting code suggestions or AWS service explanations, Q Developer.

Deciding signal

Bedrock Knowledge Bases implement retrieval-augmented generation (RAG): documents are chunked, embedded, and stored in a vector store. When a user asks a question, relevant chunks are retrieved and provided to the model as context. The model answers based on the retrieved content — it does not take actions. Bedrock Agents go further: they use foundation models to reason about a task, call tools (Lambda functions, APIs, knowledge bases), interpret results, and take sequential actions to complete a goal. When the scenario involves answering questions based on a document corpus, Knowledge Bases. When it involves an AI system that autonomously executes multi-step tasks — looking up data, calling an API, updating a record — Agents.

Deciding signal

Kendra is an enterprise search service that uses machine learning to return relevant answers from indexed documents, FAQs, and content repositories. It is traditional search with ML-enhanced relevance — keyword and semantic search over your document corpus. Lex builds conversational chatbots and voice interfaces — it handles intent recognition, slot filling, and multi-turn dialogue flows. It is the right answer when the scenario involves building a dialog interface (not just search). Amazon Q Business is a generative AI assistant that uses foundation models to synthesize answers from connected data sources — it produces natural language responses rather than a ranked list of documents. When the scenario involves building a chatbot with defined intents and utterances, Lex. When it involves a search-style interface over enterprise documents, Kendra. When it involves a generative AI assistant that synthesizes answers from company data, Q Business.

Deciding signal

SageMaker Canvas is a no-code ML interface for business analysts — it imports data, trains a model, and generates predictions through a point-and-click UI without any coding. SageMaker Autopilot runs AutoML: it automatically trains and tunes models on your data and produces an explainability report. SageMaker JumpStart provides a hub of pre-trained models that can be deployed in one click without training. The signal is user type and whether model training is happening: no-code for non-technical business users (Canvas), automated training on your own data (Autopilot), or deploying an already-trained model (JumpStart).

Deciding signal

Amazon Polly converts text to lifelike speech — given a string of text, it produces an audio file. It is used for voice-enabled applications, narration, accessibility features, and IVR prompts. Amazon Transcribe converts speech in audio or video files to text — it performs automatic speech recognition (ASR). It is used for transcription of call recordings, video captions, and voice command processing. The direction is the entire distinction: text-to-speech (Polly) or speech-to-text (Transcribe).