Skip to content

AI Engineer Roadmap

  • Roadmap: https://roadmap.sh/ai-engineer

1. Introduction

1.1 What is an AI Engineer?

  • 1.1.1 AI Engineer vs ML Engineer

1.2 Common Terminology

  • 1.2.1 AI vs AGI
  • 1.2.2 LLMs
  • 1.2.3 Inference
  • 1.2.4 Training
  • 1.2.5 Embeddings
  • 1.2.6 Vector Databases
  • 1.2.7 AI Agents
  • 1.2.8 RAG
  • 1.2.9 Prompt Engineering

  • 1.3 Impact on Product Development

  • 1.4 Roles and Responsibilities

2. Using Pre-trained Models

2.1 Pre-trained Models

  • 2.1.1 Benefits of Pre-trained Models
  • 2.1.2 Limitations and Considerations
  • 2.2.1 OpenAI Models
  • 2.2.2 Open AI Models
  • 2.2.3 Capabilities / Context Length
  • 2.2.4 Cut-off Dates / Knowledge
  • 2.2.5 Anthropic's Claude
  • 2.2.6 Google's Gemini
  • 2.2.7 Azure AI
  • 2.2.8 AWS Sagemaker
  • 2.2.9 Hugging Face Models
  • 2.2.10 Mistral AI
  • 2.2.11 Cohere
  • 2.2.12 Replicate

3. Open AI Platform

3.1 OpenAI API

  • 3.1.1 Chat Completions API
  • 3.1.2 Writing Prompts
  • 3.1.3 Maximum Tokens
  • 3.1.4 Token Counting
  • 3.1.5 Pricing Considerations
  • 3.1.6 Managing Tokens
  • 3.1.7 Open AI Playground
  • 3.1.8 Fine-tuning

4. AI Safety and Ethics

  • 4.1 Understanding AI Safety Issues
  • 4.2 Prompt Injection Attacks
  • 4.3 Bias and Fairness
  • 4.4 Security and Privacy Concerns
  • 4.5 OpenAI Moderation API
  • 4.6 Adding end-user IDs in prompts
  • 4.7 Conducting adversarial testing
  • 4.8 Robust prompt engineering
  • 4.9 Know your Customers / Usecases
  • 4.10 Constraining outputs and inputs
  • 4.11 Safety Best Practices

5. OpenSource AI

  • 5.1 Open vs Closed Source Models
  • 5.2 Popular Open Source Models

5.3 Hugging Face

  • 5.3.1 Finding Open Source Models
  • 5.3.2 Hugging Face Tasks
  • 5.3.3 Hugging Face Hub
  • 5.3.4 Using Open Source Models
  • 5.3.5 Inference SDK
  • 5.3.6 Transformers.js

5.4 Ollama

  • 5.4.1 Ollama Models
  • 5.4.2 Ollama SDK

6. Embeddings & Vector Databases

6.1 What are Embeddings

  • 6.1.1 Semantic Search
  • 6.1.2 Data Classification
  • 6.1.3 Recommendation Systems
  • 6.1.4 Anomaly Detection
  • 6.1.5 Use Cases for Embeddings

6.2 Open AI Embedding Models

  • 6.2.1 Pricing Considerations

6.3 Open AI Embeddings API

  • 6.3.1 Sentence Transformers
  • 6.3.2 Models on Hugging Face

6.4 Open-Source Embeddings

6.5 Vector Databases

  • 6.5.1 Purpose and Functionality
  • 6.5.2 Popular Vector DBs (pick one)
  • 6.5.2.1 Chroma
  • 6.5.2.2 Pinecone
  • 6.5.2.3 Weaviate
  • 6.5.2.4 FAISS
  • 6.5.2.5 LanceDB
  • 6.5.2.6 Qdrant
  • 6.5.2.7 Supabase
  • 6.5.2.8 MongoDB Atlas
  • 6.5.3 Implementing Vector Search
  • 6.5.3.1 Indexing Embeddings
  • 6.5.3.2 Performing Similarity Search

7. RAG & Implementation

7.1 RAG Usecases

  • 7.1.1 RAG vs Fine-tuning

7.2 Implementing RAG

  • 7.2.1 Chunking
  • 7.2.2 Embedding
  • 7.2.3 Vector Database
  • 7.2.4 Retrieval Process
  • 7.2.5 Generation

7.3 Ways of Implementing RAG

  • 7.3.1 Using SDKs Directly
  • 7.3.2 Langchain
  • 7.3.3 Llama Index

7.4 RAG Alternative

  • 7.4.1 Open AI Assistant API

8. Agents Usecases

8.1 Prompt Engineering

  • 8.1.1 ReAct Prompting

8.2 AI Agents

  • 8.2.1 Building AI Agents
  • 8.2.1.1 Manual Implementation
  • 8.2.1.2 OpenAI Functions / Tools
  • 8.2.1.3 OpenAI Assistant API

9. Multimodal AI

9.1 Multimodal AI Usecases

  • 9.1.1 Image Understanding
  • 9.1.2 Image Generation
  • 9.1.3 Video Understanding
  • 9.1.4 Audio Processing
  • 9.1.5 Text-to-Speech
  • 9.1.6 Speech-to-Text
  • 9.1.7 Multimodal AI Tasks

9.2 Multimodal AI APIs

  • 9.2.1 OpenAI Vision API
  • 9.2.2 DALL-E API
  • 9.2.3 Whisper API
  • 9.2.4 Hugging Face Models
  • 9.2.5 LangChain for Multimodal Apps
  • 9.2.6 LlamaIndex for Multimodal Apps
  • 9.2.7 Implementing Multimodal AI

10. AI Code Editors

  • 10.1 Code Completion Tools

11. Development Tools