Ten modules · Two hours each · Free, permanently
A free, hands-on course for senior executives. You build a working AI document-analysis system — the same architecture behind the products vendors pitch you — then you test it, measure its accuracy, and learn what it costs to run at scale.
At the keyboard: you build a document-analysis AI, plant a false fact in its sources, and watch it cite the lie back to you — then you measure its accuracy against ground truth and price it at production volume.
In the boardroom: you walk in with your own eval set, a cost model no pilot ever shows, and the questions a staged demo cannot survive.
Ten modules, self-paced. No card, ever.
You ran engineering organisations. You shipped systems on client-server, on ERP, on warehouses and networks that are still running tonight. That judgment was earned and it has not expired.
But models, embeddings, inference and agents arrived after your hands left the keyboard. So in AI conversations you have been quietly reassigned to nodding — and the worst version of it is the meeting where a board member asks an AI question and you defer to the youngest person in the room.
The gap is vintage, not aptitude. It is also smaller than anyone selling to you would like it to be. This course is not a briefing. It is the restoration of your authority to say no in your own boardroom.
We work at two altitudes at once. In the boardroom, on strategy, budget and risk. And directly alongside developers, in the code, on the systems themselves. Most people can do one. The value is in doing both, on the same problem, in the same week.
The mark is an instrument dial: sky above, ground below, and a level horizon between them. Both altitudes read on one instrument. That is the whole idea, and it is what this course hands to you.
You build an AI analyst that reads a company's board packet — the deck, the investor memo, the customer contract — and answers questions about it with citations. Then you plant a lie in those documents, watch your own system repeat it back with total confidence, and rebuild your judgment from what you saw.
What you will have built by Week 9. Every box on it is one week of work.
Everything below is free. The stack is not the curriculum — each tool exists to produce something you keep.
The workbench. Each week's build runs in your browser; nothing to install, nothing to break.
What you keep: A working system you can rerun, extend, or show anytime.
The builder. You direct it with provided prompts; it writes the code. This is how most software gets built now — directing AI rather than typing — and the coding involvement here is exactly that, at most. The prompts work in any frontier assistant — ChatGPT included. We test against Claude; use what your organization uses.
What you keep: The skill of directing AI work and judging the result, which this course pairs with the verification discipline that practice usually lacks.
The engine. Runs the AI model behind your system, with a live per-token price meter.
What you keep: A real cost model, at today's usage and at 10,000 documents.
The enterprise view. One session inside the data platform large companies actually run (Module 8).
What you keep: Recognition — you'll know what your company is buying and what the invoice pays for.
The scorecard. Your 20-question evaluation set lives in a spreadsheet.
What you keep: The scoring artifact to request from any vendor.
Set up a free environment, ingest real documents, and assemble the complete retrieval-and-generation system with cited answers.
The complete system, working, with your own hands on it — and a clear view of exactly how much a working demo does and does not prove.
Verify accuracy. Insert a known-false fact and observe the system cite it. Build a 20-question evaluation set and score against ground truth.
Retrieval misses, confident invention, stale knowledge. You cause each one on purpose, name it, and write down what true looks like.
Harden and cost it. Catch silent data errors, write a guardrail, and model the cost at production volume.
Guardrails, silent data failure, the real number at ten thousand documents. Then you reread every proposal on your desk with new eyes.
Every week ends with one sentence you can say in a meeting the next morning.
| Wk | What happens | What you can then say |
|---|---|---|
| 00 | You Will Not Spend a Dollar and You Will Not Break Anything Environment, two free accounts, first query. | “I have run this myself. Show me your version.” |
| 01 | Document Ingestion Documents in, text out. Text extraction and OCR, including scanned pages. | “Where does your pipeline lose the scanned pages?” |
| 02 | The Retrieval System Full retrieval with citations. The demo high. | “That is retrieval, not intelligence. What is it retrieving from?” |
| 03 | Accuracy Test One You insert a false fact. Your system cites it back to you. | “Your demo is only as true as the documents behind it.” |
| 04 | Failure Modes You collect your own system's three best lies. | “What does it do when the answer is not in there?” |
| 05 | Evaluation A twenty-question eval set in a spreadsheet — the artifact nobody in any vendor meeting has. | “Send me your eval set before the next meeting.” |
| 06 | Data Quality Silent data failure, reproduced on your own desk. | “Every check passed. Who noticed the number was wrong?” |
| 07 | Agents and Guardrails Agents, and the guardrail you write yourself to contain one. | “What is this agent permitted to do without a human?” |
| 08 | Cost and Model Selection Open against closed models on a live price meter. Then price it at ten thousand documents. | “That is the pilot price. Show me the number at production volume.” |
| 09 | Capstone: The System Card Capstone: your System Card. And the glossary as field guide — every term mapped to the scar that taught it. | “Here is mine. Walk me through yours.” |
Five topics, deliberately closed.
Five topics come up constantly in AI conversations, and this course deliberately leaves each one closed. You do not need what is behind it to evaluate a proposal, and pretending otherwise is how ten-week courses become eighteen-month ones nobody finishes.
We name the door each time we pass it, so you know the shape of what you are not learning. That is a different thing from not knowing it exists.
Behind the door
Two browser tabs. Two free accounts. Nothing installed, nothing to break, nothing to expense. If a week ever requires a credit card, that week is wrong and we will rewrite it.
Boardroom and keyboard, thirty years.
A career run in both rooms at once: boards, budgets and granted patents on one side; code deploying to production on enterprise data platforms this quarter on the other. The instructor is the student, thirty years in.
“I am not teaching this because I studied it. I am teaching it because it bit me, recently, at work.”
The failure in Week 6 — an FX rate that silently defaulted to 1.0 while every check passed — is not a textbook example. It is a Tuesday.
This is the first run. There are no testimonials yet — the first students are the proof. What there is: ten weeks written end to end, Week 0 through Week 9, and an instructor who will be in the thread. If that is not enough, wait for the second run. It will be better, because you will have made it better.
Self-paced · Join any time
One email to start, one per module, and at most two nudges if you go quiet — each with an off switch. Nothing else.