面向开发者的 AI
用 LLM API、提示词、RAG 和智能体进行开发。
Large language models can now write, summarize, translate, classify and even call your own code, and adding one to an app takes a single HTTP request. This track is for developers who can already program. You learn how models really work (tokens, context windows, next-token prediction and why they can be confidently wrong), call the Anthropic Messages API correctly, write prompts that return reliable JSON, stream answers and keep costs under control. Then you build real features: tool use, retrieval-augmented generation with embeddings, agents with guardrails and MCP, and the safety and testing habits that keep all of it trustworthy.
Every request is shown exactly as it travels over the network.
- 课时
- 14
- 时长
- 2 小时
- 级别
- 中级
本课程的内容目前为英文。
- 铜牌 可挑战
- 银牌 可挑战
- 金牌 未解锁
准备好证明自己了吗？
三场考试正等着你：铜牌、银牌和金牌。
学完后你将能够
- Explain tokens, context windows and sampling, and why models can be confidently wrong
- Call the Anthropic Messages API safely from a server, with streaming
- Write prompts that return JSON, and validate it before you trust it
- Control cost and limits with token budgets, caching and retries
- Build tool use, RAG and a small agent with guardrails
- Defend against prompt injection and measure quality with evals
学习旅程
1 第 1 单元Working with models
How LLMs work, your first API call, prompting, streaming and costs.
0 / 52 第 2 单元Building real features
Tool use, retrieval, agents, safety and evaluation.
0 / 53 第 3 单元Next level
Structured outputs, images and documents, deeper evals and your own MCP server.
0 / 4