LearningTree · AWS · AI & Machine Learning
AI & Machine Learning —
From Pre-Trained APIs to Custom Models
Amazon SageMaker, Bedrock, Rekognition, Comprehend, Lex, Polly, Transcribe, Translate — managed AI/ML services for every skill level, from no-code AI APIs to full model training pipelines.
01
Chapter One
Why AI/ML on AWS?
AWS AI/ML services let you add intelligence to applications without being a data scientist. Pre-trained APIs handle vision, language, and speech out of the box — while SageMaker and Bedrock give ML engineers full control over custom model training and foundation models.
The Three Layers of AWS AI/ML
AI Services (No ML Expertise)
- Pre-trained, ready-to-use APIs
- No data or training required
- Vision, speech, text, translation
- Services: Rekognition, Textract, Comprehend, Lex, Polly, Transcribe, Translate
Foundation Models (Bedrock)
- Access to LLMs from AWS + third parties
- Claude, Llama, Titan, Stable Diffusion
- Prompt engineering, RAG, fine-tuning
- Service: Amazon Bedrock
ML Platform (Full Control)
- Build, train, and deploy custom models
- Bring your own data and algorithms
- Full MLOps pipeline
- Service: Amazon SageMaker
The AI/ML Spectrum — Simplicity to Control
← No ML expertise needed Full ML control →
AI Services
Pre-trained APIs
Bedrock
Foundation models
SageMaker
Custom training
02
Chapter Two
Services
Core AI/ML Services
ML Platform
Amazon SageMaker
Fully managed ML platform — build, train, and deploy custom models at scale. Notebooks, training jobs, endpoints, and MLOps pipelines.
Deep dive →
Foundation Models
Amazon Bedrock
Access foundation models (Claude, Llama, Titan) via API. Build generative AI apps with RAG, agents, guardrails, and fine-tuning — no infrastructure to manage.
Deep dive →
AI Services
Rekognition · Textract · Comprehend · Lex · Polly · Transcribe · Translate
Pre-trained AI APIs — image analysis, document OCR, NLP, chatbots, text-to-speech, speech-to-text, and translation. No ML expertise required.
Deep dive →
Services at a Glance
| Service | Category | What It Does | ML Expertise? | Use Case |
|---|---|---|---|---|
| SageMaker | ML Platform | Build, train, deploy custom ML models | Required | Recommendation engines, fraud detection, forecasting |
| Bedrock | Foundation Models | Access LLMs via API (Claude, Llama, Titan) | Minimal | Chatbots, content generation, RAG, summarisation |
| Rekognition | Vision | Image/video analysis — faces, objects, text | None | Identity verification, content moderation |
| Textract | Document | OCR + structured data extraction | None | Invoice processing, form automation |
| Comprehend | Language | NLP — sentiment, entities, PII detection | None | Ticket routing, brand monitoring |
| Lex | Conversation | Build conversational chatbots | None | Customer support bots, IVR |
| Polly | Speech | Text to natural speech | None | Audiobooks, accessibility |
| Transcribe | Speech | Speech to text (ASR) | None | Call analytics, subtitles |
| Translate | Language | Real-time translation (75+ languages) | None | Localisation, multilingual support |
03
Chapter Three
Decision Guide
When to Use What
| If you need… | Use… |
|---|---|
| Train a custom model on your own data | SageMaker |
| Build a generative AI chatbot or content generator | Bedrock |
| Detect faces, objects, or text in images | Rekognition |
| Extract tables and forms from scanned documents | Textract |
| Analyse sentiment or detect PII in text | Comprehend |
| Build a voice or text chatbot | Lex |
| Convert text to natural-sounding speech | Polly |
| Transcribe audio/video to text | Transcribe |
| Translate text between languages | Translate |
| Combine multiple AI services in a pipeline | Step Functions + AI services |
👉 Key Takeaway
Start with pre-trained AI Services (Rekognition, Comprehend, etc.) for common tasks. Use Bedrock for generative AI. Use SageMaker only when you need full custom model training with your own data and algorithms.