Simulado AWS AI Practitioner

Cards · 58

Amazon Bedrock

Fully managed service providing access to leading Foundation Models via a single API for generative AI development.

Amazon SageMaker Studio

Integrated web-based development environment providing end-to-end tooling for ML development and deployment.

Amazon SageMaker JumpStart

Hub offering pre-trained foundation models, built-in algorithms, and pre-built ML solutions deployable in a few clicks.

Amazon SageMaker Clarify

Tool for detecting ML model bias during preparation and training, and providing explainability via feature attribution.

Amazon SageMaker Model Monitor

Continuous monitoring service that detects data drift, concept drift, and model quality degradation in production.

Amazon SageMaker Data Wrangler

Feature that simplifies data preparation, feature engineering, and data cleaning for ML using a visual interface.

Amazon SageMaker Feature Store

Purpose-built repository to store, share, retrieve, and manage ML features for both training and real-time inference.

Amazon SageMaker Ground Truth

Data labeling service combining machine learning with human labelers via Amazon Mechanical Turk or private workforces.

Amazon SageMaker Pipelines

Dedicated CI/CD workflow orchestration service built natively for automating and managing end-to-end ML workflows.

Amazon SageMaker Autopilot

Automated machine learning (AutoML) solution that builds, trains, and tunes models with full visibility and control.

Amazon SageMaker Neo

Model compiler that optimizes ML models for inference on cloud instances and edge devices without accuracy loss.

Amazon Q Business

Generative AI-powered assistant designed for enterprise use, answering questions based on corporate data repositories.

Amazon Q Developer

AI coding companion offering code generation, inline completions, security scans, and software upgrades in IDEs.

Amazon Kendra

Intelligent enterprise search service powered by natural language processing to index and query unstructured documents.

Amazon Comprehend

NLP service using ML to extract insights, sentiment, key phrases, syntax, and PII entities from unstructured text.

Amazon Comprehend Medical

Specialized HIPAA-eligible NLP service that extracts medical conditions, dosages, and PHI from unstructured text.

Amazon Rekognition

Computer vision service for automated image and video analysis, facial recognition, content moderation, and label detection.

Amazon Transcribe

Automatic speech recognition (ASR) service that converts spoken audio into accurate, punctuated text transcripts.

Amazon Polly

Text-to-speech (TTS) service using deep learning to synthesize lifelike spoken output across multiple voices and languages.

Amazon Translate

Neural machine translation service delivering fast, high-quality, and customizable text translation across languages.

Amazon Textract

ML service that extracts text, handwriting, tables, and structured form data directly from scanned documents and PDFs.

Amazon Personalize

Managed recommendation engine delivering real-time personalized user experiences based on Amazon.com ML technology.

Amazon Forecast

Time-series forecasting service using statistical and ML algorithms to predict inventory, demand, and resource metrics.

Amazon Lex

Service for building conversational AI interfaces using voice and text, powered by the same engine as Alexa.

AWS HealthScribe

HIPAA-eligible service using generative AI to create clinical summaries and transcripts from patient-doctor discussions.

Amazon Fraud Detector

Fully managed service using ML and Amazon fraud expertise to detect online identity fraud, fake accounts, and stolen cards.

Amazon CodeWhisperer

Earlier name for Amazon Q Developer, providing ML-powered code suggestions and vulnerability detection in code editors.

Amazon Bedrock Knowledge Bases

Managed RAG capability that connects Foundation Models to enterprise data stores for accurate, contextual responses.

Amazon Bedrock Agents

Autonomous orchestrators that execute multi-step business tasks by connecting foundation models to enterprise APIs.

Amazon Bedrock Guardrails

Safety feature that blocks toxic content, filters sensitive topics, and masks PII to enforce responsible AI policies.

AWS Trainium

Custom-designed AWS silicon accelerator purpose-built to deliver high-performance, cost-effective deep learning training.

AWS Inferentia

Custom AWS accelerator chip engineered to deliver high-throughput, low-latency, and low-cost ML inference.

Amazon SageMaker Model Cards

Governance tool that documents model intended use, training details, risk evaluations, and ethical considerations.

Amazon SageMaker Model Dashboard

Centralized governance portal tracking model inventory, lineage, monitor alerts, and performance metrics across accounts.

Amazon SageMaker Serverless Inference

Inference endpoint option that automatically scales compute to zero based on intermittent or unpredictable traffic.

Amazon SageMaker Real-Time Inference

Persistent endpoint configuration providing sub-second latency for workloads requiring continuous, interactive responses.

Amazon SageMaker Asynchronous Inference

Queued inference endpoint designed for large payloads up to 1 GB and long processing times up to one hour.

Amazon SageMaker Batch Transform

Offline inference option that generates predictions across non-interactive datasets and shuts down instances automatically.

Retrieval-Augmented Generation (RAG)

Generative AI architecture retrieving external enterprise data before prompting the FM to reduce hallucinations.

Temperature parameter in LLMs

Hyperparameter controlling model randomness: lower values make output deterministic, higher values increase creativity.

Top-P (Nucleus Sampling)

Inference parameter that selects tokens from the smallest candidate pool whose cumulative probability reaches P.

Top-K sampling

Inference parameter that restricts token selection strictly to the K most probable next words.

Prompt Engineering

Practice of structuring natural language input instructions, context, and examples to guide FM outputs effectively.

Fine-Tuning

Supervised process updating an existing foundation model's weights using domain-specific labeled dataset pairs.

Pre-training from scratch

Resource-intensive ML phase that initializes and trains model weights on vast unlabelled corpora to learn language patterns.

Reinforcement Learning from Human Feedback (RLHF)

Alignment technique using human preference rankings to train a reward model that fine-tunes LLM behavior.

Overfitting in ML

Condition where a model memorizes training data noise and performs poorly on unseen test data.

Underfitting in ML

Condition where a model is too simple to capture underlying patterns, showing high error on both training and test data.

Data Drift

Statistical change in input feature distributions over time compared to the data used during model training.

Concept Drift

Shift in the statistical relationship between input features and target labels over time, reducing model accuracy.

Precision metric

Proportion of true positive predictions among all instances predicted as positive; minimizes false alarms.

Recall metric

Proportion of true positive predictions captured out of all actual positive instances; minimizes missed cases.

F1-Score

Harmonic mean of precision and recall, providing a balanced evaluation metric for imbalanced class datasets.

Confusion Matrix

Evaluation table comparing actual class labels against predicted values, breaking down TP, FP, TN, and FN.

Responsible AI: Transparency

Principle ensuring stakeholders understand AI capabilities, training data lineage, limitations, and operational metrics.

Responsible AI: Fairness

Principle focused on preventing algorithmic bias and ensuring equitable model performance across diverse demographic groups.

Responsible AI: Privacy and Security

Safeguarding user data, honoring confidentiality, preventing leakage in model prompts, and enforcing IAM policies.

Amazon SageMaker Experiments

Feature that tracks, organizes, visualizes, and compares thousands of ML training runs, parameters, and results.