Skip to content

The complete learning route

From understanding the work
to owning the delivery.

A forward-deployed engineer turns an organization's difficult workflow into a useful, dependable system—and equips other people to own it. This route connects business understanding, software engineering, AI, human judgment and service operations around one project.

6 connected stages12 capability areasShared core + cloud and industry specializations
Find your starting point Developer, architect or new to software

New to software or AI

Start with the vocabulary and engineering workbench. Build and test a small program before moving into agents. Use the 101 explanations whenever a concept is unfamiliar.

Open the first principles library →

Practising developer

Use a reviewed change to establish your coding baseline. Give equal attention to discovery, evaluation, human decisions and receiving ownership—the parts a coding tool cannot decide for you.

Begin with the client workflow →

Architect or technical lead

Bring your systems experience, then demonstrate an actual implementation and recovery. Treat diagrams as claims to verify, and connect platform choices to workload and operating responsibility.

Inspect the integration proof →

Stage 3 of 6

Build and evaluate AI

Choose the simplest useful behavior, then measure what it actually does.

05

Models, agents & workflow architecture

Which behavior needs a model, and which should remain deterministic?

Understand inference and context before orchestration. Build from a bounded task to tool use, durable human approval and multi-agent communication only where the workflow warrants it.

Builds onBusiness discovery & problem framingSoftware engineering & the working environmentData foundations & trustworthy retrievalEnterprise integration & compatible change

What you need to understand

  • Model inference, token budgets and structured outputs
  • Agents versus workflows and deterministic services
  • LangChain, LangGraph, MCP and tool authority
  • Durable state, human approval and bounded delegation

Learn it in context

AI architecture foundationsBuild the mental model before the SDK
Agent engineering with LangChain & LangGraphBuild agents and stateful workflows

Course lessons use your existing training access. The practice workbook and curriculum map are free.

Evidence to bring to review

A bounded agent or workflow whose state, tools, stop conditions and human decisions can be traced and explained.

Then change one condition

A tool returns misleading content or an approval expires. Show the boundary that prevents an unintended action.

Explore an agent system step by step
06

Evaluation science & economic reasoning

How do you know it is better, and better for whom?

Measure useful outcomes with representative cases and explicit uncertainty. Separate retrieval failure, model behavior, execution and business impact instead of collapsing them into one score.

Builds onData foundations & trustworthy retrievalModels, agents & workflow architecture

Illustrative learning artifact

Inspect the result behind the score

Expected / observedSelectedWithheld
Useful draft8 correct1 missed
Unsafe draft1 unsafe pass0 rejected

8/10 correct still contains an unsafe acceptance.

What you need to understand

  • Metrics, baselines, failure slices and denominators
  • Dataset separation, provenance and label calibration
  • Retrieval measures and model-judge limitations
  • Controlled comparisons, rework and cost per accepted outcome

Learn it in context

AI evaluation & decision scienceDesign and defend a comparison

Course lessons use your existing training access. The practice workbook and curriculum map are free.

Evidence to bring to review

A versioned comparison with independent cases, failure analysis, accepted-outcome economics and a justified release recommendation.

Then change one condition

A headline score improves while a critical slice fails. Explain which use remains blocked and why.

Calculate and challenge the evaluation

Depth without losing the thread

Choose a specialization.
Keep the engineering core.

Build depth in one cloud, compare the boundaries in the others, and choose an industry context. Certification study complements the practical work; vendor exams and field acceptance are separate assessments.

Cloud & NVIDIA systemsGoogle Cloud · AWS · Azure · GPUs and private AITools & domain practiceClaude Code · Codex · LangGraph · MCP · healthcare · financial servicesReturn to first principlesPlain-language definitions, visual explanations and technical vocabulary
Why this curriculum is structured this way Research reviewed 2026-10-23

The route combines published FDE practice with cloud architecture, software delivery, reliability, AI risk, accessibility and economic reasoning. The lessons and exercises are original LockedIn Labs material. These references inform the curriculum; they do not imply vendor endorsement or accreditation.

Completion of reading or a synthetic lab does not establish field competence. The final review needs observed work, domain judgment and a receiving owner in the demonstrated environment.