Prompt Engineer Roadmap
The full map of prompt engineering from tokens to agents — 11 sections, 60+ steps.
Interactive Roadmap
Click a node → view details, access resources, mark status.
About this roadmap
An end-to-end roadmap for building production-grade LLM products. From token mechanics to agent architecture, from prompt-injection defense to eval discipline, from multimodal to multi-agent systems — **11 sections**, **60+ steps**, **150+ curated resources**. Goal: in 4-6 months, move from "someone who writes good prompts" to **an engineer who designs and maintains production-grade LLM products**.
Who is it for?
Junior/mid software engineers, data scientists, ML engineers; anyone moving from 'using' to 'building with' LLMs.
What you'll learn
- Develop intuition for modern LLM internals (transformer, attention, decoding) and cost dynamics
- Consciously tune token budgets, context windows, and sampling parameters
- Write production-grade prompts using zero-shot, few-shot, role + delimiter patterns
- Solve hard problems with CoT, ToT, ReAct, Reflexion, Skeleton-of-Thought
- Guarantee 100% structured output with Pydantic/Zod schema + function calling + streaming
- Optimize cost and accuracy with prompt caching, long-context, hybrid retrieval + reranking
- Apply RAG, GraphRAG, episodic memory patterns in production
- Build multimodal LLM integrations: vision, document, audio, video
- Master tool-use, agentic loops, MCP servers, computer use, and multi-agent orchestration
- Understand the prompt patterns powering agentic IDEs like Claude Code and Cursor
- Design eval datasets, run LLM-as-judge, version prompts, A/B test, and monitor costs
- Layered defense against prompt injection (direct + indirect), jailbreak, and hallucination
- Compliance-ready production with PII detection, content moderation, and red-teaming
- Integrate GDPR, KVKK, EU AI Act requirements into AI product design
- Specialization patterns across coding, SQL, support, legal, healthcare verticals
- Career progression from junior to senior via portfolio, open-source contribution, interview prep
Starting context
Basic Python/JS, REST API concepts, 10+ hours playing with any LLM playground. ML/deep-learning background not required.