Book 3 of 6 · Level 03 · Advanced · 03 · Toleranced
Level 3 · Advanced — Primwright: AI Engineering & Production Systems
An Advanced Guide to Engineering, Securing, Deploying, Observing, and Optimizing AI Systems
The advanced level: production AI as an engineering discipline. Advanced means system responsibility — typed interfaces, explicit state, measured trade-offs, and a defensible reliability case.
Engineer → Secure → Deploy → Observe → Optimize → Scale
The working shape of the level, in the subtitle's own verbs: engineering first, then securing what you built, deploying it, observing it, optimizing it — and scaling only what has proven itself.
What the level covers.
- Typed, replaceable interfaces and explicit state
- Model mechanics tied to measurable engineering consequences
- Hybrid retrieval, reranking, faithfulness, and abstention
- Evaluation harnesses, threat models, telemetry, rollback, and incident response
- Measured latency, throughput, quality, safety, and cost trade-offs
Reserved for later specialization — not in this level — not in this level.
- Research-level machine-learning mathematics
- Foundation-model and distributed GPU training
- CUDA or kernel engineering
- Advanced reinforcement learning
- Deep MLOps specialization and advanced AI-security research
The weeks, in order.
- Week 1 · Production AI System Architecture
- Week 2 · LLMs Under the Hood
- Week 3 · Model Selection, Routing, and Adaptation
- Week 4 · Advanced Retrieval-Augmented Generation
- Week 5 · Agents, Orchestration, and Memory Engineering
- Week 6 · Evaluation Engineering and the Harness Gate
- Week 7 · Security Engineering and Red-Team Gate
- Week 8 · Fine-Tuning and Adaptation
- Week 9 · Infrastructure, Deployment, and LLMOps
- Week 10 · Performance, Cost, and Scale
- Week 11 · Capstone Construction
- Week 12 · Red Team, Reliability Case, and Defense
Who it's for.
Engineers who ship AI systems and own the consequences — latency, cost, safety, and failure. If Book 2 was about building AI applications, Book 3 is about being responsible for them at production scale.
Prerequisites.
- Book 2 competencies — building, connecting, evaluating, and securing AI applications. Each level assumes the last; none can be skipped.
- An entry diagnostic — the 90-minute reliability repair — protects the level: a document Q&A service with four faults (an unbounded request, a malformed schema path, one retrieval miss, and a tool without authorization). Passing requires safe repairs, one regression test, and residual risk named; learners who cannot modify the system complete bridge work before Week 1.
What you get.
Every level pairs a book with a course. The book is the curriculum — the complete text and the standard of evidence. The course is the guided path through it — sessions, knowledge checks, and capstone work.
The book.
Primwright: AI Engineering & Production Systems — advanced guide to engineering, securing, deploying, observing, and optimizing ai systems at Level 03 (Advanced).
No standalone book page yet — this level page carries the level's record until release.
The course.
Course 3 is the guided companion to Book 3 — sessions, knowledge checks, and capstone work. Format and access are announced with the release.
Availability, stated plainly.
After Level 03: AI Systems Architecture & Leadership.
Level 04 · Architect — An Expert Guide to Designing, Governing, Evolving, and Leading Enterprise AI Systems