←
Momchil Penev80 posts
Momchil Penev avatar

Momchil Penev

@vedrinaM

building skillia.ai to make skills transfer effortless, scalable and visual

◷ Joined August 2013▧ 13 Photos and videos
Momchil Penev avatar
Momchil Penev@vedrinaM·Nov 5

skillia.AI is now on Microsoft Marketplace. Expert records the job once. A voice guide trains new operators on it. AI checks every attempt step by step. The skills matrix keeps itself current. Buy on your Microsoft account. Link below.

Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 15

I went deeper on how this works on a real shop floor in this thread 👇

Let's surface this thread here. The initial question was "How do you actually get AI to help me on my shop floor?" 1/5 You can record a video of your best operator and use AI to create a detailed step-by-step SOP with images, guidance and focus points

Post attachment
Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 15

Step 3: keep a live skills matrix of who has proven what. Pay and scheduling follow proven competencies, not tenure, and single points of failure show up before someone quits or retires.

0:15
Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 15

Step 2: the expert demonstrates each process once on video. That becomes the standard. Trainees are recorded doing the real work, AI scores each step against it, and the expert reviews only the edge cases.

0:15
Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 15

Great piece by @daniipreneur. The second half is where most plants get stuck. Here's how I'd operationalize it. Step 1: define "qualified" as competencies you can actually observe on the job, not hours in a class.

0:15

Two simple ways to solve the Talent Shortage! Thanks to friends at @anduriltech, @saronic, and @spacex for the insights.

Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 9

A first-article report proves the part. It doesn't show whether the operators on the next shift follow the approved method. We wrote the long version: what operator competency evidence should look like when production ramps, and where it sits next to AS9102, PPAP and audits. Link below.

Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 7

Amazon plans to have 20,000+ smart glasses on its delivery drivers by the end of 2027: navigation, package checks, hazard alerts, and the proof-of-delivery photo, all hands-free. That's proof the package arrived. Manufacturing needs the next step: proof the person did the job the way the expert demonstrated it. Guidance helps on the shift. Evidence is what lets you sign someone off.

Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 7

A first-article report proves the part. During a ramp, the other question is whether each supplier's operators were trained on the approved method and follow it. I wrote up what that evidence should look like.

Momchil Penev @vedrinaM Oct 6 Article Proof of the part. Proof of the person. What the supplier base needs to show when production ramps. On October 2, a16z published Connor Love and Collen Larson's essay on the American manufacturing supplier base. Their point, briefly:

Momchil Penev avatar
Momchil Penev@vedrinaM·Oct 2

Manufacturing Month talk keeps landing on "more training." The gap we keep seeing is proof: after the class or the ride-along, can you show who is actually cleared on that SOP? skillia.AI captures the expert once, trains people hands-free on the approved standard, then proves it on camera. We don't spend the expert's time on every training and every assessment.

Momchil Penev avatar
Momchil Penev@vedrinaM·Sep 23

A signed training record still gets treated like proof someone can run the station. Usually it just proves they sat through something. It says nothing about whether they can handle that job at 2am when the usual person is out. We keep confusing exposure with capability — and plants pay for that mix-up in scrap, overtime, and near-misses that everyone somehow saw coming.

Momchil Penev avatar
Momchil Penev@vedrinaM·Sep 23

Plants keep asking how to “deploy AI on the shop floor.” Most answers jump straight to copilots and dashboards. The useful version is smaller. Film your best operator once. Turn that into standard work people can actually follow. Then have the next person run the job on camera and score them against those steps — not against whether they sat through training. That’s deploy. Everything else is theater until you know who can still run the station when the expert is out. If you’ve tried putting AI on a line, what broke first — the docs, the changeover, or the handoff?

Momchil Penev avatar
Momchil Penev@vedrinaM·Sep 8

Multiple times now I've watched plants treat a signed training record like proof someone can do the job. All it really proves is that they sat through something. It says nothing about whether they can run that station at 2am when the usual person is out. We keep confusing exposure with capability, and you pay for that mix-up in scrap, overtime, and near-misses that everyone somehow saw coming.

Momchil Penev avatar
Momchil Penev@vedrinaM·Jun 30

today we had our first customer demo using the newly integrated @MentraGlass . amazing results and new glasses ordered for the customer on the spot @caydengineer & team, you made us look like geniuses, thank you!