The catalog

The path teaches you to build AI systems. The catalog teaches you to govern them, advise on them, and prove it with a finished project. Same method: every lesson teaches, coaches, has a lab you run, and ends with a check.

Courses

• Course · 4 lessons · about 2.9 h

  AI Governance

  Governance as engineering work. You write Pocket's inventory record and tier it with a rule anyone can rerun, score a risk register before and after the guardrails and evals you already built, turn an acceptable-use policy into checks that run over every trace, and end with a one-page report for leadership: a status from a written rule, four metrics, the top risks and one clear ask.

  For: Developers who build or run LLM features and need to answer a security lead, an auditor or leadership with evidence rather than assurances.

• Course · 4 lessons · about 3 h

  AI Security

  Security as engineering work on a real agent. You map Pocket's attack surface as a graph and find every path from untrusted text to a way out, attack it with a red-team corpus scored per category and gated in CI, stop secrets and system prompts from leaking through answers, logs and rendered links, and end by pinning the tools you depend on, writing detection rules over your traces and rehearsing the response.

  For: Developers who build or run LLM features with tools or retrieval and need to show that the system holds up against prompt injection, leakage and a compromised dependency.

• Course · 4 lessons · about 2.8 h

  AI Consulting

  Advise a company on rolling out Pocket-style assistants. You score its readiness across data, people, process, technology and governance and find the binding constraint, rank candidate use cases on value and feasibility with risk as a filter, build a business case from a measured baseline with NPV, payback, a sensitivity table and break-even adoption, and compare building against buying with a weighted matrix, three-year TCO and a phased roadmap gated by exit criteria.

  For: Developers who have built an LLM app and are now asked whether, where and how their company should adopt AI.

Capstones

Multi-sitting projects. Each ends with a rubric you score your own work against, and a walkthrough of what strong work looks like.

• Capstone · about 12 h

  Train a text classifier from scratch

  Route Pocket's notes into topics with a model you train yourself, and prove it beats a baseline.

• Capstone · about 16 h

  Ship a documentation assistant with evals

  Point Pocket's retrieval stack at a new corpus, measure it end to end, and make a launch call on evidence.

• Capstone · about 14 h

  Build an agent that works within a budget

  A research-and-summarize agent on Pocket with hard limits, guards, 15 graded tasks and a cost-versus-quality report.

• Capstone · about 16 h

  From readiness to roadmap: an AI engagement

  Run a full engagement for a fictional software company, from interviews to a one-page memo leadership can sign.