AI 102
Maps Azure AI services, Vision, Language, OpenAI, Document Intelligence, with links, code snippets, and provisioning details for a Portuguese-language workshop.
What this file does
Maps Azure AI services, Vision, Language, OpenAI, Document Intelligence, with links, code snippets, and provisioning details for a Portuguese-language workshop.
When to use it
- Studying for AI-102 certification with Portuguese materials
- Building a workshop curriculum on Azure AI services
- Quickly referencing Azure AI service capabilities and endpoints
- Learning how to provision and call Azure AI Vision and Language APIs
Assumes this stack
AI 102
Sessão 01
- Conceito sobre Data Science link
- Conceito sobre Inteligência Artifical link
- Conceito sobre Machine Learning
link
- Responsible AI
link
- Fairness
- Reliability and Safety
- Privacy and Security
- Inclusiveness
- Transparency
- Accountability
- Azure AI Services
link
- Capabilities
- Language
- Text Analysis
- Question Answering
- Language Understanding
- Translation
- Speech
- Speech Recognition
- Speech Synthesis
- Speech Translation
- Speaker Recognition
- Vision
- Image and Video analysis
- Image Classification
- Object detection
- Optical character recognition
- Generative
- Generate text completions
- Image Generation
- Language
- Provisioning
link
- Models
- Multi-Service Azure AI Services
- Single Service
- Containers (need Internet)
- Endpoints
- Key
- Location (Regions)
- REST API
- SDK
- Models
- AI Services
- Azure AI Vision
link
- Image Analysis
link
- Caption and Tag generation
- Object Detection
- People Detection
- Optical Character recognition
- Smart Crop Thumbnail
- Background removal
- Multi-modal Embeddings
- Product recognition
- Method Analyze() CODE
- Inputs
- File or URL
- Visual Feature
- VisualFeatures.Caption
- VisualFeatures.DenseCaptions
- VisualFeatures.Tags
- VisualFeatures.Objects
- VisualFeatures.SmartCrops
- VisualFeatures.People
- VisualFeatures.Read
- Analysis Options (optional)
- Cropping aspect ratios
- Gender Neutral caption T/F
- Language em/pt
- Model Version
- Output
- captionResult -> {"text":value, "confidence":value}
- objectResult -> ["text":value, "confidence":value]
- smartCropsResult -> [{"boundingbox": {"x":value,"y":value,"w":value,"h":value}}]
- Read Feature -> ["lines":[{"text":value, "boundingPolygon":{"x":value,"y":value}}, "words": {"text":value, "boundPolygon"{"x":value,"y":value}, "confidence":value} ]]
- Inputs
- Face Detection
link
- Face Service
- Face Detection CODE
- Comprehensive facial feature analysis
- Face attribute analysis
- Facial landmark location (nose, mouth)
- Facial liveness
- Face comparison and identification (require access approval)
- Data privacy and security
- Transparency
- Fairness and Inclusiveness
- Facial recognition (require access approval)
- Data privacy and security
- Transparency
- Fairness and Inclusiveness
- IMPORTANT: Toda face identificada recebe um id anônimo (24 horas de duração)
- Persisted face Recognition
- Add person group
- Add person
- Persisted Faces
- Add person
- Add person group
- Inputs
- Options
- returnFaceId, returnFaceLandmarks, returnFaceAttributes, recognitionModel
- Options
- Outputs
- "faceId"
- "recognitionModel"
- "faceRectangle"
- "faceLandmarks": {"pupilleft":{"x":value, "y":value}}
- Face Service
- Video Indexer
link
CODE
- Facial Recognition
- Optical character recognition
- Speech transcription
- Topics
- Sentiment
- Labels
- Content Moderation
- Scene Segmentation
- Custom Insights
- Facial Recognition (celebrities)
- Language - industry/organization specific terms
- Brands
- Image Analysis
link
- Azure AI Vision
link
- Capabilities
- Responsible AI
link
Sessão 02
- Conceito sobre Inteligência Artifical
- Azure AI Services
- AI Services
- Azure AI Language
link
- Analyze Text
link
- Language Detection - Detecta o idioma
- Key Phrase Extraction - Detecta as palavras chave de um texto
- Sentiment Analysis - Detecta sentimentos (neuro, positivo, negativo, misto) com uma determinada confiança
- Name Entity Recognition - Categoriza entidades (pessoa, local, data e hora, organização, endereço, e-mail, url)
- Entity Linking - usa Wikipedia como base de dados para definir substantivos
- Summarization
- Extractive summarization - Frases mais importantes
- Abstractive summarization - Produz o resumo com a ideia principal
- Personally Identifiable Information Detection (PII)
- Detectar e remover dados sensíveis
- Translating Text
link
- Detection - ISO Language Code
- Translation - Traduz para idiomas solicitados como path param
- Transliteration - Tradução fonética entre idiomas
- Custom Translation - Modelo customizado de tradução para domínios específicos link
- Question Answering
link
- Responde perguntas através de uma base de conhecimento
- Language Understanding
link
- Natural Language Processing (NLP) - interface que suporta linguagem Natural
- Entrada em linguagem natural, modelo determina a intenção, aplicativo reliza a ação
- Utterance - enunciado
- Que horas são? Me fale as horas
- Intent - intenção
- Saber a hora
- Entities - entidades
- Prebuilt entities: (quantidade, data e hora, e-mail, telefone, URL)
- Custom text Classification
- Classifica documentos de texto em classes
- Custom Named Entity Recognition
- Classifica documentos de texto para encontrar entidades
- Speech
- Speech-to-text - duas APIs (Short Audio API)
- Text-to-speech - duas APIs (Batch synthesis API)
- Speech Synthesis Markup Language (SSML) - estilos, pausas, enfase, audio de fundo
- Speech Translation API
- Speaker Recognition API
- Natural Language Processing (NLP) - interface que suporta linguagem Natural
- Analyze Text
link
- Azure AI Language
link
- AI Services
- Azure AI Services
Sessão 03
- Conceito sobre Inteligência Artifical
- Azure AI Services
- Azure OpenAI
link
- Generative AI - modelos com capacidade de gerar textos, imagens, audio e vídeo através de um prompt
- Azure OpenAI resource
- Azure AI Studio
link
- Catálogo de modelos
- GPT-4 / GPT-3 / Embeddings / DALL-E / Phi / Llama / Mistral / Hugging Face
- Benchmarks
- Fine Tuning
- Prompt Catalog
- Playground
- Habilidades: classificação de sentimento, geração de conteúdo, tradução/transformação, sumarização, continuação de texto, resposta de perguntas, chat
- Integração com aplicações
- Endpoint + chave
- Azure OpenAI SDKs
- Semantic Kernel link
- Catálogo de modelos
- Prompt Engineering - técnicas para definir as melhores formas de interagir com os LLM
link
- Instruções claras
- Grounding Content
- Dicas sobre como responder
- System Messages
- Conversation History
- Chain of thoughts - pedir para o modelo responder passo a passo
- Retrieval Augmented Generation (RAG)
- Azure OpenAI - criar uma fonte de dados, contectar na fonte, usar o modelo
- Knowledge Mining Solution
link
- Intelligent Search
- Azure AI Search - indexa documentos, busca dados
- Data Source / Skillset (data enrichement) / Indexer (mapeamento de dados) / Index (coleção de dados consultável)
- Custom Skill
- Knowledge Store - Guarda dados sobre os dados (relacionamentos, estrutura dos documentos)
- Azure AI Search - indexa documentos, busca dados
- Intelligent Search
- Azure AI Document
link
- Extração de dados de documentos (OCR)
- Modelos prontos ou customizados para seu formulário
- Azure OpenAI
link
- Azure AI Services
What's inside
3 sessions, 6 Azure AI services, 12 feature lists, 4 code examples, 30+ links
Change this for your project
- Replace
wt.mc_id=AZ-MVP-5003638with your own Microsoft MVP tracking ID - Replace
pt-brin URLs with your target locale if not Brazilian Portuguese - Replace
[CODE](...)links with your own code samples or remove them
Where it goes
Reference documentation for a retrieval pipeline. Keep with the ingestion or retrieval code it describes.
Worth borrowing
- Organizing AI service capabilities into a hierarchical outline for easy scanning
- Pairing each service with a direct link to official Microsoft Learn documentation
- Including code snippet references inline with feature descriptions
Related Documents
SUMMARY
Proposes three on-prem AI architectures, modular, hybrid, and fully local RAG, with hardware specs and vendor lists.
Retrieval & Prompts
Explains how CharMemory's extraction prompt and Vector Storage settings determine memory retrieval quality in SillyTavern.
Advanced RAG Techniques
Surveys advanced RAG architectures, Agentic, Corrective, Self-RAG, Graph RAG, RAPTOR, Multimodal, Temporal, with design trade-offs and decision guidance.
RFC-BLite: High-Performance Embedded Document Database for .NET
Specifies an embedded document database for.NET with zero-allocation I/O, C-BSON format, and ACID transactions.