00.3 — Complete Roadmap

The 17-part structure of this handbook

 1. Software Engineering Foundation        (Python, APIs, FastAPI, SQL, Postgres, Redis, Git, Linux, Testing)
 2. AI/ML Fundamentals                     (ML, DL, NLP, Transformers, Embeddings, LLMs)
 3. AI Engineering                         (Prompting, structured output, tool calling, RAG, agents, MCP, memory, multimodal)
 4. LangChain                              (architecture, internals, production patterns)
 5. LangGraph                              (state machines for agents, persistence, HITL, multi-agent)
 6. LangSmith                              (tracing, evals, monitoring)
 7. Production AI Engineering              (Docker, CI/CD, cloud, K8s, scaling, queues, gateways, cost)
 8. AI Reliability / GenAIOps              (evals, observability, guardrails, red teaming)
 9. AI Security                            (prompt injection, RBAC, tenant isolation, secure tool execution)
10. Enterprise Integration                 (OAuth/SSO, enterprise DBs, CRM/ERP, governance)
11. System Design for AI                   (RAG/agent/multi-tenant architectures from first principles)
12. Customer Discovery and FDE Skills      (requirements, feasibility, communicating to non-technical stakeholders)
13. Rapid Prototyping                      (FDE engagement methodology, demos)
14. Deployment in Customer Environments    (on-prem, air-gapped, VPC, customer IT constraints)
15. Business Impact and ROI                (measuring what actually matters to the customer)
16. End-to-End FDE Projects                (8 full projects, one simulating a complete engagement)
17. FDE Interview and System Design Prep

Why this order (and how it differs from the original whiteboard roadmap)

The original roadmap that inspired this book placed System Design after Enterprise Integration. This handbook moves System Design (Part 11) before the deepest enterprise integration and before the FDE customer-facing layer, for a specific reason: you need to understand why an architecture looks the way it does — where the queue goes, why there's a cache, why retrieval is separate from generation — before you can reason about how to bolt a customer's SSO, CRM, and compliance requirements onto it. Learning enterprise integration first, without system design, produces engineers who can wire up an OAuth flow but can't tell you why the RAG pipeline needs a reranking step or where a multi-tenant system will fall over at 10x load.

The other deliberate addition is Data Engineering, folded into Part 3 (embeddings/vector DBs), Part 10 (enterprise pipelines), and the projects in Part 16, rather than given its own top-level part. Reasoning: in real enterprise AI work, the LLM call is rarely the hard part. Getting a customer's documents, databases, APIs, permissions, and business processes into a reliable pipeline is usually the majority of the engineering effort — and this is exactly the kind of thing tutorials skip and customers judge you on.

If you are already a working AI engineer

Your highest-leverage path through this book is not linear front-to-back. It is:

AI Engineering (Part 3, focus on internals you've skipped)
   → Production AI (Part 7)
   → System Design for AI (Part 11)
   → GenAIOps / Security (Parts 8–9)
   → Enterprise Integration (Part 10)
   → FDE skills (Parts 12–15)
   → Projects (Part 16)
   → Interview prep (Part 17)

Parts 1–2 and 4–6 are still worth reading in full — this book assumes you know LangChain/LangGraph/LangSmith at a basic level and spends its pages on architecture and production patterns you likely haven't seen, not on syntax you already have. But if you are extremely pressed for time, those are the parts you could skim first and return to.

A note on scope creep in real FDE roles

There is a meaningful overlap between AI FDE work and security engineering: prompt-injection defense, RBAC, authorization boundaries, data privacy, API security, and secure agent/tool execution show up constantly in enterprise AI deployments, because giving an LLM agent access to real company systems is, structurally, giving a somewhat unpredictable program access to real company systems. Part 9 (AI Security) is written with that weight — it is not an afterthought bolted onto the end of the AI engineering material.

Every part should end with something built, not just read

The material is written so each part accumulates toward the Part 16 projects. If you only read and never build, you will be able to discuss the architecture in an interview and fail to actually deliver it in the first month of a real FDE role. Do the exercises.