Complete Learning Cycle Β· 12-17 weeks total

The 5-Phase Learning Path

A carefully designed path validated by hundreds of learners. Each phase has clear inputs, outputs, and validation criteria. Follow it and you'll go from completely not understanding to personally delivering a complete large-scale Agent project.

Complete Timeline

A clear path

Bottom to top, each phase has a starting point, endpoint, and validation method.

Phase 1 Β· 1-2 weeks

Cognition & Mental Models

Understand what an Agent is, why, and how. Build a complete mental map, master LLM, prompts, reasoning, and function calling.

πŸ“š 7 core concepts πŸ§ͺ 5 mini labs βœ… 6 self-check items
πŸ“₯ Output

Can explain Agent essence to anyone. Complete the 6-item self-check independently.

Phase 2 Β· 2-3 weeks

Skills & Tool Stack

Master mainstream frameworks (LangChain, LangGraph, CrewAI, AutoGen, LlamaIndex). Learn to select, combine, and customize.

πŸ“¦ 5 frameworks πŸ”§ 20+ tool integrations πŸ“Š 3 comparison tables
πŸ“₯ Output

Can choose the most suitable framework combo for any requirement with rationale.

Phase 3 Β· 3-4 weeks

Complete Project Build

Personally assemble discrete modules into an engineering project. From architecture to testing to deployment ops.

πŸ›  4 layered modules πŸ§ͺ Complete test suite πŸ“¦ CI/CD pipeline
πŸ“₯ Output

A runnable complete Agent system with docs, tests, and visual interface.

Phase 4 Β· 2 weeks

Evaluation & Optimization

Build eval systems, master monitoring. Learn prompt optimization, cost control, error analysis.

πŸ“Š 10+ eval metrics πŸ’° Cost optimization πŸ› Debug methodology
πŸ“₯ Output

Complete eval pipeline + monitoring dashboard + optimization iteration log.

Phase 5 Β· 4-6 weeks

Independent Large-Scale Agent

Synthesize everything, from requirements to delivery, build a complete large-scale Agent yourself.

🎯 Custom requirements πŸ— Complete architecture πŸš€ Deployment
πŸ† End

A complete, production-grade, open-source-able Agent project.
It's your best work.

Phase Details

Each phase in detail

Choose to learn in timeline order or pick the phases you need most.

🧠 Phase 1 · Cognition 1-2 weeks

  • 1.1 LLM fundamentals
  • 1.2 Prompt Engineering core
  • 1.3 Function Calling
  • 1.4 Agent paradigms overview
  • 1.5 Reasoning & CoT
  • 1.6 Memory & retrieval
  • 1.7 Evaluation & observability

πŸ”§ Phase 2 Β· Frameworks 2-3 weeks

  • 2.1 LangChain & LCEL
  • 2.2 LangGraph stateful composition
  • 2.3 LlamaIndex RAG expert
  • 2.4 CrewAI multi-Agent
  • 2.5 AutoGen dialogue framework
  • 2.6 Advanced: build your own engine

πŸ›  Phase 3 Β· Build 3-4 weeks

  • Config management
  • Modular design
  • Observability
  • Error handling
  • Test suite
  • Deployment

βš™οΈ Phase 4 Β· Evaluation 2 weeks

  • Eval datasets & benchmarks
  • Performance & cost monitoring
  • Prompt & parameter tuning
  • Retrieval optimization
  • Context compression
  • Caching strategies
  • Parallelization
  • Reflection loops

🎯 Phase 5 · Delivery 4-6 weeks

  • Design from scratch
  • Modular code organization
  • Tests, docs, CI/CD
  • Deploy & operate

πŸ“¦ Project Templates

Pick a direction that solves a real problem for you:

  • πŸ›’ Smart shopping assistant
  • πŸ“š Company knowledge base Agent
  • πŸ” Automated research assistant
  • πŸ’Ό Sales automation
  • πŸ’» Coding Agent
  • πŸŽ“ Education tutor

Start with step one

Cognitive foundation is bedrock. This phase doesn't need coding, but needs continuous thinking and experimentation.

Phase 1 Β· Build Cognition β†’