AI for developers: APIs and agents

About this course

This course is for developers who want to move from toy scripts calling an LLM API to production code — with proper error handling, cost control, and predictable behavior under load. It addresses the problem of brittle integrations that break under real traffic: no retries, no limits, no tests. The course brings an engineering approach to working with language models.

The topics are real engineering: a client with timeouts, retries, and backoff; RAG with chunking, embeddings, hybrid search, and cited generation; function calling and an agent loop defended against prompt injection; monitoring and caching. Projects have you build a production-ready wrapper around an LLM API, a working RAG pipeline over real documents, an agent with tools, and finish by bringing one real LLM feature to production readiness.

This is the developer track and assumes programming experience in any language. No prior experience with LLM APIs specifically is needed — the course starts from the basics of a model request and tokens.

What you'll learn

• Write an LLM API client with timeouts, retries with backoff, and error handling
• Estimate the token cost of a scenario across different models
• Build RAG: chunking, embeddings, hybrid search, reranking, and cited generation
• Implement function calling and an agent loop with a step limit
• Defend agent tools against prompt injection
• Write evals for prompts and test changes before shipping
• Set up monitoring, observability, and request caching
• Design system evolution: versioning, A/B tests, model updates

What you'll end up with

• A production-ready LLM API wrapper with tests, retries, and error handling
• A working RAG pipeline over real documents
• An agent with at least three tools and action logging
• An architecture diagram for a production-ready LLM feature, plus a certificate on paid plans

This course isn't for people who don't write code — it's an engineering course built on API examples, not a tour of AI tools for end users.