Level 1: Beginner

Core concepts, models & prompting

Level 2: Intermediate

Data layer, embeddings & RAG

Level 3: Advanced

Agents, MCP & Multimodal

AI Engineer Roadmap

Based on roadmap.sh/ai-engineer

Builds applications on top of pre-trained models & APIs
(distinct from ML Engineers who train models from scratch)

Function Calling & Structured Output

  • Function Calling concepts
  • Tool use patterns
  • Structured output formats (JSON, schemas)

AI Agents

  • What are AI Agents?
  • Agent Use Cases
  • ReAct Prompting

Frameworks:

  • OpenAI AgentKit & Agent SDK
  • Claude Agent SDK
  • Manual Implementation

Model Context Protocol (MCP)

  • MCP Host / Server / Client
  • Data Layer & Transport Layer
  • Building an MCP Server
  • Building an MCP Client
  • Connect to Local Server

Multimodal AI

  • Image Understanding
  • Image Generation (DALL-E)
  • Video Understanding
  • Audio Processing
  • Text-to-Speech / Speech-to-Text

Tools: OpenAI Vision API, Whisper API, HuggingFace Models, LangChain/LlamaIndex multimodal

Core LLM Concepts

  • Large Language Models (LLMs)
  • Tokens & Context Window
  • Inference vs Training
  • AI vs AGI
  • Prompt Engineering (overview)
  • Vector DBs & RAG (overview)

Embeddings

  • What are Embedding
  • Semantic Search
  • Recommendation Systems
  • Anomaly Detection
  • Data Classification

Providers: OpenAI, Cohere, Gemini, Jina, Sentence Transformers, HuggingFace

Pre-trained Models

  • Anthropic Claude
  • OpenAI GPT / o-series
  • Google Gemini
  • Meta Llama (open source)
  • Mistral (open source)
  • Cohere

Prompt Engineering

  • Zero-Shot Prompting
  • Few-Shot Prompting
  • Chain of Thought (CoT)
  • ReAct (Reasoning + Acting)
  • Top-K / Top-P Sampling
  • Robust prompt design

Closed Source APIs

  • OpenAI Response API
  • Google Gemini API

Vector Databases

  • Indexing Embeddings
  • Similarity Search

Options:
Chroma, Pinecone, Weaviate, FAISS, LanceDB, Qdrant, Supabase, MongoDB Atlas

RAG (Retrieval-Augmented Generation)

  • What is RAG? / Use Cases
  • RAG vs Fine-tuning

Pipeline:
Chunking → Embedding → Vector DB → Retrieval → Generation

Frameworks: LangChain, LlamaIndex, Haystack, RAGFlow

AI Safety & Ethics

  • Prompt Injection Attacks
  • Bias & Fairness
  • Security & Privacy Concerns
  • Adversarial Testing
  • Content Moderation APIs
  • Add end-user IDs in prompts
  • Constrain inputs & outputs
  • Know your customers / use cases

Introduction

  • What is an AI Engineer?
  • AI Engineer vs ML Engineer
  • Roles & Responsibilities
  • Impact on Product Development

Open Source Tools

  • Ollama (run models locally)
  • LM Studio
  • Hugging Face Hub
  • Hugging Face Inference SDK
  • Transformers.js