tech-studies

AI Engineer

2 min read

AI Engineer

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

Suggested path through the AI Engineer nodes. Each node links to its lesson when written.

Nodes

Fundamentals

  • What is an AI Engineer (vs ML Engineer)
  • How LLMs work (inference vs training)
  • AI vs AGI
  • Impact on product development
  • Roles & responsibilities

LLM basics

  • Large Language Models (LLMs)
  • Tokens & tokenization
  • Context window & context engineering
  • Sampling params — temperature, top-p, top-k, repetition penalty
  • Pre-trained models — benefits & limitations
  • RAG vs fine-tuning — when to use which

Models & providers

  • Closed vs open-source models
  • OpenAI GPT / o-series
  • Anthropic Claude
  • Google Gemini
  • Open models — Llama, Mistral, DeepSeek, Qwen, Gemma, Cohere
  • Choosing the right model
  • Self-hosting — Ollama, LM Studio, Hugging Face Hub
  • OpenRouter & OpenAI-compatible APIs

Provider APIs

  • OpenAI Responses / Chat API
  • Claude Messages API
  • Google Gemini API
  • Streaming responses
  • Structured output
  • Tools / function calling
  • Prompt caching
  • Using SDKs directly

Prompt engineering

  • Prompt engineering basics & system prompting
  • Zero-shot & few-shot
  • Chain of thought (CoT)
  • ReAct prompting
  • Role & behavior shaping
  • Prompt injection attacks
  • Robust / adversarial prompting

Embeddings

  • What are embeddings
  • Embedding models (OpenAI, Gemini, Cohere, Jina, sentence-transformers)
  • Indexing embeddings
  • Semantic / similarity search

Vector databases

  • Vector databases — Pinecone, Weaviate, Qdrant, Chroma, FAISS, pgvector

RAG

  • What is RAG & use-cases
  • Chunking
  • Retrieval process & dynamic filters
  • Implementing RAG (LangChain, LlamaIndex, Haystack)

AI agents

  • AI agents & use-cases
  • Tools & function calling
  • ReAct & multi-agents
  • External memory
  • Model Context Protocol (MCP) — host, client, server
  • Agent SDKs — OpenAI AgentKit, Claude Agent SDK, Google ADK

Multimodal

  • Multimodal AI & use-cases
  • Image understanding & generation (DALL·E, vision API)
  • Video understanding
  • Audio — text-to-speech & speech-to-text (Whisper)

Dev tools

  • Frameworks — LangChain, LlamaIndex, Haystack
  • AI coding tools — Cursor, Windsurf, Replit, Codex, Claude Code

Safety & ops

  • AI safety & ethics
  • Bias & fairness
  • Security & privacy concerns
  • Adversarial testing & content moderation
  • Fine-tuning (when, LoRA)

Resources

See resources.md.

Project ideas

  • Build a RAG chatbot over your own notes (chunk → embed → retrieve → answer)
  • Build a tool-calling agent (MCP server + client) for a real task
  • Ship a multimodal feature (image-in → structured-out) via a provider vision API

1 item under this folder.