AI Exam 4 - Products

AI Practitioner Exam Prep - Products

Abbreviations

KMS = Amazon Key Mgmt Service product; A2I = Augmented AI

Developer Products (in ML frameworks layer)

SageMaker AI = IDE plugin that is fully managed service that automates ML lifecycle (from data prep to production) with "no-code" environment and handles infrastructure to streamline building, tuning, and deploying models. Capabilities: Predictive analytics, computer vision, NLP, and fraud detection.

Operations: Features auto-training and integrated logging. Security & Data: Secured via IAM; integrates with S3 (storage), Lambda (triggers), CloudWatch (monitoring) and API Gateway (endpoints).

    Auto Model Tuning feature exists on Sagemaker.

    Auto Pilot = uses Clarify to show how ML models could make predictions. uses SHAP values. Auto finds the best hyperparameters.

    Canvas = No-code ML tool to create predictions. Has AutoML piece that is focused on models.

    Clarify = evaluates FMs for accuracy, bias, robustness, and toxicity. Helps with transparency and explainability.

    Code Editor = connects to VS Code

    Data Wrangler = rebalances data for undersampling, oversampling or systemic minority oversampling.

    forecasting algorithms:

       parameters: ARIMA, DeepAR+ (time-series), ETS, and LSTNet. 

       non-parameters: CNN-QR (predicts quadriles), NPTS, and Prophet (time-series).

    Debugger = debugs the code

    Endpoints = fully managed service for ML. via HTTPS URLs. allow your apps to send data to ML model and receive a prediction.

    Feature Store = stores and shares the features/variables of a model to team.

    Ground Truth = data labeling service creates high-quality training datasets by sending the most difficult ones (hardest 30%) to crowd-sourced humans by outsourcing data labeling tasks. Has special computer vision labeling section.

     HyperPod = managed infrastructure service for accelerated distributed training and fine-tuning of FMs. 

     Inference or Deployment options = deploys trained models as hosted services with 4 options: 1) Real-Time: Good for low-latency, but has low payload. 2) Serverless: Good for intermittent traffic with idle periods. 3) Asynchronous: Good for large payloads or long-running. 4) Batch Transform: Good for offline massive datasets. Works for high-volume, scheduled.

    JumpStart = Gen AI. ML hub with 100s of FMs and pre-built MLs (vision, NLP, and tabular data) deployable with a few clicks. Not a low code option. Created code should feature threat detection and data protection. Some of evaluation types such as automatic model evaluation. Can restrict FMs.

    MLOps  = DevOps for ML

    Model Dashboard = sharing team info on production model behavior in one place.

      Model Cards = your model property info for documentation purposes

      Model Registry = store, manage, tracks your model versions and through deploy ML lifecycle.

    Model Invocation Logging = you can turn this on.

    Model Monitor = monitors production models for data drift, model drift, and quality loss.

    Purchase Provisioned Throughput = dedicated capacity bought. required for using custom models.

    Role Manager = define min. permissions

    Studio = web-based IDE for ML with tools for data prep, model building, training, and deployment.

    Studio notebooks = Jupyter notebooks with ML libraries and tools. Use Studio notebooks to write, run, and share code for data exploration, model training, and deployment.

    Training = trains ML models on various compute instances, including GPU accelerated instances using distributed service.

AI/ML Products (in services layer)

Augmented AI =  builds workflows for human review allowing adjust confidence levels

Comprehend = NLP (text). calcs unlabeled pre-trained toxicity detection, redaction, sentiment, entities, key phrases, and topics NLP models.

Kendra = NLP (text). enterprise search service to find context or semantic searches. 

Lex = Think Lexicon.  Conversational brain for speech and text to apps. Interaction focused. Such as chatbot, voice controlled menu, powers Alexia, etc. Outputs intent.

OpenSearch Service = Store embeddings in a vector database via the scalable index mgmt and nearest neighbor search, then later search via keyword and NL matching. Good for semantic searches and similarity-based recommendations since embeddings. 

Personalize recommends. Ex: movie recommendations. Only recommendations and no other purpose.

Polly converts text into speech/audio. Think Polly Anna Parrot. WaveNet is the GenAI part of it.  Polly itself is not used to build the dialogue flow or understand user input.

Rekognition = auto image and video analysis for your apps without ML experience. Think eyes. Outputs labels, text, and data.

   Medical extracts structured medical information (such as medication, condition, test results) from unstructured text. 

Textract extracts typed and written text from cursive or scanned typed document images. No search.

Transcribe converts speech into text. 

Translate is a text translation to different language.

Gen AI Products (in services layer)

Bedrock = fully managed container service for FMs from 3rd parties (such as AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon) or custom for GenAI. Has IAM security (with custom roles that you can limit data access), watermark detection, single API, and good S3 link. If fetch error via S3, then check decryption rights. Can create a RAG agent to fetch data in it.

    Pricing: 1) On-Demand = good for variable or unpredictable workloads. 2) Provisioned Throughput: good for steady workloads or custom models. Users commit to a set throughput for 1 or 6-month periods for lower costs. 3) Batch Processing: good for asynchronous, large-scale tasks. 

    Agents = enables AI assistants to interact with specific data sources, query external APIs, create and compare responses, and prioritize results. Good for chatbots, claims, etc.

    Endpoints = fully managed. can even limit to region.

    Guardrails = filters bad or PII user input to FMs. can filter bad topics. good for children content.

    Model Evaluation = compares the FMs giving metrics. 

    Nova Canvas model = creates/edits hi-res image from prompt

    Nova Lite model = supports multiple languages and is low cost.

    Nova Pro model = text/image/video analysis

    Nova Reel model = creates videos

    PartyRock = very cheap and experimental environment for learning about gen AI apps, allowing users to quickly build, test, and iterate on AI apps without incurring significant costs.

    Stable Diffusion 3.5 Large = creates high quality images based on text inputs.

    Titan model = creates/edits low res image from prompt

Q Business - BI. Answers questions using your company data.

Q Developer - Helps developers with test case creation, documentation creation, code recommendations, opens source license tracking, reference tracking, and snippets. Works with Glue.

QuickSight - fully managed, cloud-native BI. Connects to S3, RDS, Redshift, on-premises dbs, and Salesforce, GitHub.

Security and Compliance Products in AI

Artifacts = Think museum artifacts. Portal to see 1) compliance reports 2) and agreements. No cost. Compliance reports are ISO, PCI, SOC 1, SOC 2, and BSI C5. Emails compliance and security reports on ISVs and other third parties on updates.

Audit Manager = continually audits your usage, compliance with industry regulations and internal policies by creating audit reports.

Cloud Trail logs tracking user activity and API usage even on stopped EC2 instances. No UI. Helps with governance, compliance, and operational and risk auditing.

Config audits configurations of resources. Managed. Monitors config changes for compliance, security, and change mgmt. Works in the control tower's landing zone.  

Guard Duty =  24/7 intelligent threat detection across your infrastructure and resources. Agentless. Creates security logs. Monitors behavior (attacking).

Inspector = Monitors vulnerabilities, for storage of EC2 and containers, and Lambda functions. Like building inspector.

Key Mgmt Service = Create and manage crypto keys. AWS (same account) and customer managed keys (for cross-account). Think physical key.

Macie = monitors PII and sensitive info and secures data (data in S3) at rest. Uses ML.

Security Hub = Overall state of security and compliance. Not security vulnerability.

Shield = Auto. 

      Standard = Protects against external common security issues

      Advanced = Protects against external DDOS

Trusted Advisor monitors real-time cost, performance, resilience, security, and service quotas. 

Misc Products

A2I = human reviews of ML predictions

AI Service Cards = use cases and limitations, etc. on various AI services.

Aurora - PostgreSQL using pgvector= good to store vector db. good for vector similarity searches and integration with structured data.

DeepRacer = 1/8 race car demo code to learn RL.

EC2 = chips with lowest environmental cost are EC2 Trn (trainium) chips.

Fraud Detector = fully managed. uses ML to find payment fraud, fake account creation, or policy abuse. Only fraud checking and no other purpose.

HealthScribe = HIPPA clinical app that analyzes doctor/patient conversations.

Lookout for Metrics = obsolete. monitors for anomalies in business metrics

OpenSearch Service = stores vector databases (such as custom ML models) as a fully managed service that supports vector data types, for storing and querying embeddings efficiently. Has scalable index mgmt and nearest neighbor searches which are good for vector db. used in AI similarity searches.

Privacy Reference Architecture = guidelines for privacy