AI Intermediate

Build and deploy AI applications using Azure AI Foundry hub, Azure OpenAI Service, and Prompt Flow.

• Azure AI Foundry: hubs, projects, connections, model catalog
• Azure OpenAI Service: deployments, quota, model versions (GPT-4o, o1, etc.)
• Azure OpenAI endpoints: completions, chat completions, embeddings, DALL-E
• Prompt Flow: DAG flows, LLM nodes, Python nodes, evaluation flows
• Azure AI Search with Azure OpenAI: integrated vectorization, semantic ranking
• Content Safety: harm categories, severity levels, blocklists
• Managed online endpoints: deploying models, traffic split, blue-green
• Azure AI Studio vs Foundry: understanding the rebrand and migration
• Azure OpenAI on your data: bring your own data with AI Search
• Monitoring AI applications: Azure Monitor, token usage dashboards, latency tracking

Free Azure AI Foundry & Azure OpenAI practice questions →Azure AI Foundry & Azure OpenAI study guide →

Use Vertex AI Model Garden, Gemini API, and Google AI pipelines for building production AI solutions.

• Vertex AI overview: Model Garden, Workbench, Pipelines, Matching Engine
• Gemini on Vertex AI: gemini-1.5-pro, gemini-1.5-flash, gemini-2.0, context caching
• Vertex AI Gemini API vs Google AI Studio: when to use each
• Grounding with Google Search: reducing hallucinations, citation sources
• Vertex AI Agent Builder: search apps, conversational agents, data stores
• Vertex AI evaluation: pointwise vs pairwise, eval datasets, Gemini as judge
• Vertex AI Pipelines: KFP-based, component authoring, pipeline runs
• Multimodal inputs: video, audio, image with Gemini on Vertex AI
• Tuning on Vertex AI: supervised fine-tuning, RLHF, distillation
• IAM for Vertex AI: service accounts, workload identity, roles

Free Google Vertex AI & Gemini practice questions →Google Vertex AI & Gemini study guide →

Build AI applications with Amazon Bedrock foundation models, Agents for Bedrock, and AWS AI services.

• Amazon Bedrock overview: FMs (Claude, Titan, Llama, Mistral, Stability AI)
• Bedrock API: InvokeModel, Converse API, streaming, cross-region inference
• Knowledge Bases for Bedrock: S3 data source, OpenSearch Serverless, retrieval
• Agents for Bedrock: action groups, Lambda functions, OpenAPI schema
• Bedrock Guardrails: content filters, PII redaction, denied topics, grounding
• Model evaluation on Bedrock: automatic vs human, task types, metrics
• Amazon SageMaker vs Bedrock: when to use each
• AWS AI services: Rekognition, Comprehend, Textract, Transcribe, Translate
• Bedrock pricing: on-demand vs provisioned throughput, batch inference
• Security on Bedrock: VPC endpoints, KMS encryption, CloudTrail logging

Free AWS Bedrock & AI Services practice questions →AWS Bedrock & AI Services study guide →

Use Hugging Face Hub, the transformers library, and open-source models including local inference with Ollama.

• Hugging Face Hub: model cards, datasets, spaces, organizations
• Model licenses: Apache 2.0, MIT, Llama Community License, RAIL — usage rights
• Hugging Face Inference API: hosted inference, authentication, rate limits
• transformers library: pipeline(), AutoModel, AutoTokenizer, from_pretrained
• Model quantization: GGUF, GPTQ, AWQ, BitsAndBytes — tradeoffs
• Ollama: running models locally, ollama pull/run, Modelfile, REST API
• Popular open models: Llama 3.x, Mistral, Phi-3, Qwen, DeepSeek — use cases
• Sentence Transformers: embedding models, semantic similarity, bi-encoders
• Hugging Face Spaces: Gradio and Streamlit demos, GPU Spaces
• Fine-tuning basics: LoRA, QLoRA, datasets format, training scripts

Free Hugging Face & Open Source Models practice questions →Hugging Face & Open Source Models study guide →

Evaluate LLM quality systematically and apply fine-tuning techniques to adapt models to specific tasks.

• Evaluation dimensions: faithfulness, relevance, coherence, groundedness
• Automated evals: ROUGE, BLEU, BERTScore — limitations with generation tasks
• LLM-as-judge: using GPT-4/Claude to score outputs, consistency bias
• RAGAS: context recall, faithfulness, answer relevance metrics for RAG
• Evaluation datasets: creating golden datasets, annotation guidelines
• When to fine-tune vs prompt engineer: task specificity, data volume, cost
• Supervised fine-tuning (SFT): dataset format (prompt-completion pairs), chat format
• LoRA and QLoRA: low-rank adaptation, rank/alpha hyperparameters, VRAM savings
• RLHF and DPO: preference learning, reward models, direct preference optimization
• Fine-tuning platforms: OpenAI fine-tuning API, Vertex AI tuning, Hugging Face AutoTrain

Free LLM Evaluation & Fine-tuning practice questions →LLM Evaluation & Fine-tuning study guide →