How Jordan and Sam built their own compliance training platform with AI, using Veo 3.1 scenario videos, Supabase and Vite React instead of buying software.
Yes, AI can build a full compliance training platform, including the training videos. On Episode 20 of Built This Week, Jordan Metzner and Sam Nadler showed the internal tool they built for their own company: AI wrote the curriculum, AI coded the platform, and every scenario video was generated with Google's Veo 3.1. Jordan's view is that the build cost little beyond development time, while comparable software is sold per user, per month.
Jordan says most people have sat through compliance training on sexual harassment or phishing, and that in big companies it is legally required. Their company needed it for certain compliance work, and he thinks a yearly training for everybody is good practice anyway. His complaint about the usual platforms is that they feel like 1990s videos of awkward workplace scenes followed by a quiz. He saw AI as a way to generate the content, generate the exam platform and build the whole thing end to end.
The admin view he demoed has tests, a training hub, certificates, user management and settings, in dark and light mode. A certificate builder lets admins pick borders, fonts and wording, save templates and apply them to different exams. Each exam is 10 questions, and each question should take 30 to 90 seconds. The screen shows a passing score of 70, but Jordan says he set it up so you have to get 100, and a wrong answer means you answer again. The full build is written up on our AI compliance training platform page.
Watch the AI compliance training platform demo in Episode 20 (starts at 3:15) · Episode 20 notes and transcript
Every question starts with a scenario video, and users must watch at least 95% of it before they can answer. Jordan says the videos were made 100% with AI, using Veo 3.1 fast as far as he recalls, and that every video in the platform was made the same way. In his view you would not easily notice they are AI.
The example on the show: a coworker comments on someone's dress in the break room, and the question asks how to respond. The options are to ignore it, flirt back, politely set boundaries and report if the behavior continues, or laugh it off. Sam declined to joke about it and picked setting boundaries and reporting, which was the correct answer.
Jordan says these education platforms are expensive and priced per user, usually somewhere from $10 to $100 per user per month in his estimate. Their version plugs into the rest of their technology stack, and besides development time he puts the cost at zero. It has a small back end on Supabase, a Vite React front end, and Veo for the test content.
He also searched Google for sexual harassment training software on air to make a point: page after page of sponsored ads and vendors. His takeaway:
This is part of this new world of AI kind of build versus buy where this case, the cost to build was pretty low and the cost to buy is high, and we're gonna be able to retain control of the software and the product over the long term and continue to make it better for our particular use case.
— Jordan Metzner, Episode 20
Sam says the choice was buy it or build it, and the build used AI in three pillars. First, AI organized the curriculum: which tests to run, which questions go in each test and how the training meets the requirement. Second, AI helped build the platform itself. Third, Veo 3.1 produced the content. It took a few passes, and Sam estimates 60% to 70% of the content was right the first time. He also noted this one took longer than a day or so to build, unlike some of the quicker tools on earlier episodes.
He thinks the idea travels well beyond compliance. A company could build an orientation platform the same way, with short videos about life at the company and a certificate at the end.
I love when we build something that anyone who listens to this podcast could take the idea and be like, oh, I could do that for my organization.
— Sam Nadler, Episode 20
The tool of the week was Pomelli, Google's social marketing tool. Sam explains that you enter your website and it builds a "business DNA" from it: images, fonts, brand colors, copy and taglines. You can edit it or rerun it, but for him it was right the first time. From there it creates campaigns in seconds.
Jordan ran it on the site for Ntrvsta, his AI interview product, which we covered in Episode 6. Reading the site takes a few minutes, and it asked him for the logo. It then suggested campaigns such as "recruitment speed simplified," which Sam liked as a tagline even though he didn't love the images. Jordan asked it for four new LinkedIn images, used "fix layout" to get new versions and had it suggest fresh campaign ideas. Some results looked very AI, and the output is static images only, though Jordan imagines Google adding Veo 3.1 to it. Sam says his teams have started using it for quick inspiration and high quality content.
In the news segment, Sam brought up Amazon testing robot warehouses in Whole Foods so shoppers can pick up other orders at checkout. Jordan, who worked at Amazon for years, explained the background. Fulfillment centers used to have people walking aisles with pushcarts. After Amazon bought Kiva, shelves started coming to a picker who stays still. He imagines a Whole Foods where the shelves move to the picker instead of someone walking the store, though he wasn't sure that is how this test works.
His bet is that robots will make grocery orders faster, cheaper and easier, so going to the supermarket becomes something people choose rather than have to do. Sam's angle was the shopper problem: his family buys some things at Whole Foods and the rest at other grocery stores every week. If Whole Foods can hand over items it doesn't stock, those extra trips go away. More on this topic is on our AI in retail and ecommerce page.
Sam says the Neo home robot was priced at about $20,000 up front or around $500 a month. Jordan says these launches are often too good to be true, and points to a Wall Street Journal tester who found a human on the other side controlling it. That makes the robot an interface for a person, not yet more efficient. He believes Tesla also had humans controlling Optimus at some launches.
Having a humanoid robot do all these things in your house is a dream. I think it's a very difficult reality.
— Jordan Metzner, Episode 20
His reasoning: driving has a small set of controls, while doing dishes changes with every house, sink and dishwasher. On price, he thinks $20,000 is right, roughly half the price of the cheapest Tesla Model 3, which he puts at maybe $40,000. Sam thinks $500 a month is fair because monthly home services such as a weekly cleaner easily cost more. But he believes we are a long way from it working, mechanically and because of how many variables a home has compared with autonomous driving, which people have discussed for ten years.
Jordan walks through the certificate builder, the exam flow and a live Pomelli run on camera. Watch Episode 20 of Built This Week, where the full transcript is also on the page.
