Appearance
Generative AI Training Plan — Source of Trust
Source: Google Doc:
[Huy Chau] Generative AI Training Plan
Doc ID:1E80LeQMymp0Yu88JlNP0MV_6Kva1fISlIw5jXNuPAh0
Author: Huy Chau
Created/Updated: September 16, 2026
Workspace:/Users/huychau/Documents/working/training/ai
🧭 Plan Structure & Navigation
| File | Phase / Topic | Duration | Key Focus |
|---|---|---|---|
| 00_master_plan.md | Master Curriculum & Overview | 10+ Weeks | Architecture, Objectives, Tech-Stack, Roadmap |
| 01_phase1_engine_and_prompting.md | Phase 1: The Engine & Programmatic Prompting | 1 Week | Model Knobs, Prompt Architecture, LangFuse, Injection Defense |
| 02_phase2_engineering_of_reasoning.md | Phase 2: The Engineering of Reasoning | 1 Week | CoT, Chain of Draft, ToT, ReAct, AlignmentCheck |
| 03_phase3_modular_chaining_lcel.md | Phase 3: Modular Chaining & LCEL | 1 Week | Runnable Protocol, LCEL, Pydantic Structured Outputs, Fallbacks |
| 04_phase4_context_engineering_agentic_rag.md | Phase 4: Context Engineering & Agentic RAG | 1.5 Weeks | 3-Layer Memory, Agentic RAG, GraphRAG, Ragas, BadRAG Defense |
| 05_phase5_harness_engineering_mcp.md | Phase 5: Harness Engineering & MCP Consumption | 1.5 Weeks | Scaffolding, FastMCP/MCP, Build-Verify-Fix Loops, AIRateLimiter |
| 06_phase6_stateful_graphs_hitl.md | Phase 6: Stateful Graphs & Human-in-the-Loop | 2 Weeks | LangGraph v1, DCGs, TypedDict State, Checkpointing, HITL Gates |
| 07_capstone_nexus_deep_researcher.md | Capstone: NEXUS (The Deep Agent) | 3.5 Weeks | Deep Researcher (VFS, Reasoning Sandwich), Federated Swarm (A2A, L9) |
| 08_production_evaluation_engineering.md | Production Evaluation Engineering | 1 Week | LLM-as-a-Judge, Annotation Queues, Automated Data Flywheel |
| 09_beyond_industrialization_ecosystems.md | Beyond: Industrialization & Ecosystems | 2 Weeks | Guardrails, IoA, CA-MCP, Token Economics, Redis Caching, Streaming |
| 10_rapid_development_deepagents.md | DeepAgents: Rapid Development Track | 1 Week | create_deep_agent, write_todos, VFS, Context Isolation |
🛠 Technology Stack & Tools
- Orchestration: LangGraph v1 (Stateful Graph & Functional API), LangChain (LCEL)
- Protocol: Model Context Protocol (MCP), Agent-to-Agent (A2A) Protocol
- Evaluation: RAGAS, DeepEval, Promptfoo, LLM-as-a-judge
- Observability: LangFuse (Tracing, Trajectory Evals, Runs/Threads), LangSmith
- Security: Meta LlamaFirewall (CodeShield, PromptGuard 2), LLM Guard, AWS Bedrock Guardrails, OWASP Top 10 for Agentic Apps (2026)
- Engines & Storage: ChromaDB, Qdrant, Redis (Semantic Caching), vLLM