Skip to main content

Learning Path

Phase 1: Build core LLM application skill​

Start with provider abstractions, prompt contracts, streaming, and structured outputs. The main outcome is an API layer that can talk to multiple models while keeping response formats stable.

Phase 2: Add retrieval and grounded answers​

Learn how document pipelines, chunking, embeddings, retrieval, reranking, and citations turn a chat endpoint into a knowledge system.

Phase 3: Move from single calls to orchestration​

Add planning, state, tools, multi-step execution, retries, and approvals. This is the move from “chat feature” to “agent workflow.”

Phase 4: Productize the operator experience​

Expose plans, timelines, approvals, sources, and execution state inside a serious UI. This is where many demos fail and real products begin.

Phase 5: Harden for production​

Security, evals, observability, CI/CD, and deployment are first-class topics because enterprise adoption depends on them.

Completion criteria​

You are done when you can explain and demonstrate:

  • one production-ready provider integration
  • one grounded RAG workflow
  • one multi-step tool-using orchestrator
  • one operator-facing UI with approvals
  • one release process with observability and deployment documentation