ERW CommandPMO

AI-103 CERTIFICATION BOOTCAMP

Developing AI Apps and Agents on Azure — Build it, run it, name it exactly.

40 Sessions · 45 Minutes Each · 8 Weeks at Your Own Pace · One track: your own Azure subscription
Enrolment open  ·  self-paced · no cohort dates · one payment of $199.99 · every assignment marked, and reviewed by R. Wingo if it falls short
Coach Ritchie, the ERW AI assistant
Welcome

Hello — I am Coach Ritchie, the AI assistant for this bootcamp. R. Wingo wrote the course, and your work is marked against the rubric he published; anything that falls below the mark to continue, he reviews himself, because that is the point at which a person needs help rather than a score. I am here for the moments in between, when a command will not run or two services look the same.

40 sessions, 45 minutes each, over 8 weeks. AI-103 is the exam that asks whether you can build AI apps and agents in Python on Microsoft Foundry and run them properly: choose the model and the service, deploy it, connect to it without a key, ground it with retrieval, build agents with tools, watch them with tracing, keep them safe and within quota, and extract meaning from documents, images, text and speech. You will write and run real code in your own subscription, and every assignment is marked on the evidence that it ran.

Every assignment comes back as a written letter: what you got right, and where something fell short, exactly where to look. You will never get a bare score. Ask me anything as you work — paste a screenshot of the error if that is quicker — but I will not do the assignment for you, and I switch myself off while you sit a quiz or an exam.

Coach RitchieERW AI assistant · the course is written by R. Wingo, who reviews every assignment marked below the pass

Start here

$199.99one payment · the whole course · every assignment marked · the marked Final · certificate and reference letter · Coach Ritchie · three questions a day to R. Wingo
1 · BUY

Your purchase email carries a licence key. Keep it.

2 · ENROL

Sign in with Google, give the name for your certificate, and enter the key.

3 · SESSION 1

Choosing the model: LLMs, small language models, multimodal models, Foundry Tools.

SESSION 40

The capstone and the marked Final. Pass both and the certificate is yours.

Buy the AI-103 bootcamp — $199.99 →

Already bought? Enrol with your key →  ·  Enrolled? My progress →  ·  Just looking? Session 1 is open to read now, and so is every other session. →

What AI-103 measures

Microsoft publishes the skills it measures and how heavily each area is weighted. The sessions follow that outline bullet by bullet — a check in the build proves every official skill is taught by a session — and the time spent on each area sits inside Microsoft’s band.

Skill areaWeight
APlan and manage an Azure AI solution25–30%
BImplement generative AI and agentic solutions30–35%
CImplement computer vision solutions10–15%
DImplement text analysis solutions10–15%
EImplement information extraction solutions10–15%

Source: Microsoft’s official study guide, skills measured as of April 16, 2026. Read the study guide on Microsoft Learn → If Microsoft changes the outline, the course is updated to match.

One track: your own Azure subscription

AI-103 is an associate exam about building AI apps and agents in Python on Microsoft Foundry. It measures things that only exist in a real subscription: model and agent deployments, managed identity, private networking and keyless credentials, quotas and rate limits, Azure AI Search indexes and skills, Content Understanding analyzers, Translator and Speech, image and video generation, and Foundry projects in CI/CD pipelines.

Said plainly: this course requires your own Azure subscription with a payment method. You will run real deployments and pay for what you use. Every session tells you what it will create, whether it is billed, and exactly how to delete it, and Session 1 has you set a budget alert before anything is deployed. There is no no-card track for AI-103, because a student on one would reach the exam unable to pass it.

You also need Python on your own machine and to be comfortable writing and running a short script. If that is new to you, start with a Python course first.

The names, exactly

Every session ends its teaching with a box of the exact names the exam uses — services, tiers, roles, portal blades, commands, SDK packages, classes, methods and parameters — each one linked to the Microsoft Learn page that defines it, and each one checked against that page when the session was written. All of them are collected on one printable cheat sheet.

Download the AI-103 cheat sheet (PDF)

The marker holds you to the same standard: where an answer is about a service or an API, a correct description with the name missing or wrong earns at most half of that criterion.

What happens when you submit

STEP 1You submit

One document per session, Word or Google Docs, with a Session: N line at the top and your screenshots pasted in.

STEP 2It is marked

Against that session’s published rubric, criterion by criterion, with your screenshots read as evidence.

STEP 3A letter comes back

What holds up, what fell short, and where to look — never the answer itself. Usually within the hour.

STEP 4Resubmit if you need to

Below 70, fix it and send it again. A resubmission carries a flat five-mark deduction. Above 85 a session is closed — move on.

Evidence beats description. Where a criterion depends on something you ran or configured, full marks need the evidence — the output, the values, the screenshot. A confident description of what would happen is capped at 70% of that criterion.

The 40 sessions

8 weeks, five sessions a week, 45 minutes each, at whatever pace your diary allows. Every session is open to read before you buy. Every session ends in one assignment, marked and returned as a letter.

WEEK 1

Plan the solution and write the first code

Open Week 1 →
  1. 1Choosing the model: LLMs, small language models, multimodal models, Foundry ToolsUpload session 1
  2. 2Choosing the Foundry service, and the retrieval and indexing methodUpload session 2
  3. 3Designing the Azure infrastructure, and choosing deployment optionsUpload session 3
  4. 4Hands-on: configuring model and agent deploymentsUpload session 4
  5. 5First code: connect an app to a Foundry project and call deployed models · POP QUIZ 1Upload session 5
WEEK 2

Generative AI inside applications

Open Week 2 →
  1. 6Tuning generation: prompt engineering and model parametersUpload session 6
  2. 7Retrieval-augmented generation (RAG) in an applicationUpload session 7
  3. 8Workflows, tool-augmented flows and multistep reasoningUpload session 8
  4. 9Evaluating models and apps: fabrications, relevance, quality, safetyUpload session 9
  5. 10Integrating generative workflows with the Foundry SDKs and connectors · POP QUIZ 2Upload session 10
WEEK 3

Agents

Open Week 3 →
  1. 11Choosing memory, tool and knowledge services for agentsUpload session 11
  2. 12Defining an agent: role, goals, conversation tracking, tool schemasUpload session 12
  3. 13Agents with retrieval, function calling and conversation memoryUpload session 13
  4. 14Agent tools: APIs, knowledge stores, search, Content Understanding, custom functionsUpload session 14
  5. 15Multi-agent orchestration, and orchestrating models, flows and rules engines · POP QUIZ 3Upload session 15
WEEK 4

Running agents: autonomy, observability, quotas and the mid-course review

Open Week 4 →
  1. 16Autonomy with safeguards and approvals; reflection and self-critique loopsUpload session 16
  2. 17Observability: tracing, token analytics, safety signals, latency; agent error analysisUpload session 17
  3. 18Quotas, scaling, rate limits and cost footprintsUpload session 18
  4. 19Monitoring models and data: drift, safety events, grounding, index health, relevanceUpload session 19
  5. 20Mid-course review: every name so far, and practice exam 1 · PRACTICE EXAM 1Upload session 20
WEEK 5

Security, pipelines, responsible AI and ingestion

Open Week 5 →
  1. 21Security: managed identity, private networking, keyless credentials, role policiesUpload session 21
  2. 22Foundry projects in CI/CD pipelinesUpload session 22
  3. 23Responsible AI: safety filters, guardrails, risk detection, moderation, evaluatorsUpload session 23
  4. 24Auditing and governing agents: trace logs, provenance, approvals, tool-access controlsUpload session 24
  5. 25Ingesting and indexing documents, images, audio and video; the RAG ingestion flow with OCR · POP QUIZ 4Upload session 25
WEEK 6

Search, extraction and visual understanding

Open Week 6 →
  1. 26Semantic, hybrid and vector search; enrichment with skillsUpload session 26
  2. 27Wiring retrieval into workflows and agents; multimodal extraction pipelinesUpload session 27
  3. 28Content Understanding for documents: grounded representations and analyzersUpload session 28
  4. 29Content Understanding for images and video: pipelines, segments, objects and regionsUpload session 29
  5. 30Multimodal understanding: visual context, captions, visual Q&A, alt-text · POP QUIZ 5Upload session 30
WEEK 7

Generating images and video; analysing text

Open Week 7 →
  1. 31Image generation and editing: reference media, inpainting, masks, controlsUpload session 31
  2. 32Video generation and editing generated videoUpload session 32
  3. 33Responsible AI for images: unsafe content, injected text in images, visual policyUpload session 33
  4. 34Extracting entities, topics, summaries and structured JSONUpload session 34
  5. 35Sentiment, tone, safety and sensitive content; domain-specific outputs · POP QUIZ 6Upload session 35
WEEK 8

Translation, speech and the capstone

Open Week 8 →
  1. 36Translation: Azure Translator in Foundry Tools vs LLM translation flowsUpload session 36
  2. 37Speech for agents: speech to text, text to speech, custom speech modelsUpload session 37
  3. 38Reasoning over audio, and speech translationUpload session 38
  4. 39Every name on one page: full-outline drill and practice exam 2 · PRACTICE EXAM 2Upload session 39
  5. 40Capstone review and exam technique · THE MARKED FINALUpload session 40

Assessments

8 practice papers, marked the moment you submit, every question returned with the reasoning behind it, and a marked Final that works the way the real exam is weighted. The session that opens each paper says so and flags itself on its tab, and your progress page always names the paper due now. Closed book, no notes, and Coach Ritchie switches himself off while you sit one. Practice papers: pass mark 70, resit as often as you like.

Pop Quiz 1 — Planning and First Code

8 questions · 12 minutes · after Session 5

Pop Quiz 2 — Generative Applications

8 questions · 12 minutes · after Session 10

Pop Quiz 3 — Agents

8 questions · 12 minutes · after Session 15

Practice Exam 1 — Mid-Course

40 questions · 80 minutes · after Session 20

Pop Quiz 4 — Security and Ingestion

8 questions · 12 minutes · after Session 25

Pop Quiz 5 — Search and Understanding

8 questions · 12 minutes · after Session 30

Pop Quiz 6 — Generation and Text

8 questions · 12 minutes · after Session 35

Practice Exam 2 — Full Mock

50 questions · 100 minutes · after Session 39

The Marked Final

50 questions · 100 minutes · in Session 40 · marked on the server, out of 1000, pass 700

The marked Final is not a practice paper. It is served by the server with the answers removed, drawn fresh at every sitting in proportion to Microsoft’s published domain weights, and marked once on the server. It is scored out of 1000 and you need 700. No answers come back — the letter names the sessions to revisit. Resit it as often as you need; the best score counts.

When you finish: the certificate, the letter, and a gift

What you receiveWhen
Certificate of completionWhen your Session 40 capstone is marked at 70 or above and your marked Final is 700 or above.
Reference letter from R. WingoWith the certificate, naming the work you did.
The Intelligent Barrel — the full book — and its three bonus PDFs
  • The Reader’s Toolkit
  • The 90-Day AI Pilot Charter & Playbook
  • The Value–Risk Map & Autonomy Playbook
A gift for finishing: the claim link is emailed to you when your capstone passes.
One thing to know about the gift. The claim code only works for existing customers, and Gumroad recognises you by the email address you check out with. Use the same address you bought this course with.

This is an ERW CommandPMO credential, not a Microsoft certification. The Microsoft certification is Microsoft Certified: Azure AI Apps and Agents Developer Associate, which you earn by booking and passing AI-103 with Microsoft yourself.

House rules

Questions people ask first

QuestionAnswer
Can I read before I buy?Yes. Every session is open to read. Buying gets your work marked, the marked Final, Coach Ritchie, questions to R. Wingo, the certificate and the gift.
How much time, honestly?45 minutes per session plus about an hour for each assignment. Call it 60 to 80 hours over eight weeks.
Does it include the Microsoft exam fee?No. You book AI-103 with Microsoft yourself and pay Microsoft’s fee. The last session shows you where.
Will it cost me anything in Azure?Yes, a modest amount if you delete as you go. Every session says what it creates and how to remove it, and you set a budget alert before anything is deployed. Check current prices on Microsoft’s pricing pages.
What if I miss a week?Nothing happens. There are no cohort dates and no deadlines.
Who marks it?R. Wingo, against the published rubric. Coach Ritchie helps you think; he does not mark, and he does not do your assignment.
Is there a guarantee I will pass AI-103?No honest course can promise that. What this course promises is that every skill Microsoft lists is taught, every name is checked against Microsoft Learn, every assignment is marked against a published rubric, and R. Wingo reviews any assignment that falls below the mark to continue.

Read reviews from students who have passed →