Letting Engineers Engineer, While We Handle the Paperwork

Too much of nuclear knowledge is locked away in aging documents, fading memories, and burdensome workarounds. At Everstar, we’re building AI tools that don’t just search or summarize—they understand, evaluate, and draft like your best engineer would. It’s time to stop reinventing the wheel and start unlocking what we already know.

At Everstar, we know nuclear, but we still listen first. We’ve spoken to hundreds of industry insiders about how they do their work and what frustrates them the most. Though the outputs might be different, it’s all congruent with what I saw over two decades at the U.S. Nuclear Regulatory Commission. This gives us the conviction to build solutions that will provide immediate relief and value. To face the pressures they’re under—delivering more outcomes (usually power), safely, with fewer or less-experienced staff—the time has never been better for AI adoption in this industry.

The Big Picture Is Clear

When we zoom out, we realize that people are only really doing three things in pursuit of licenses, approvals, or continued safe operation:

* finding the right information—the needle in the nuclear haystack

* thinking and deciding whether something is safe, acceptable, etc.—whatever the criteria

* documenting that decision—for the regulator, for your boss, in a press release

The Old Ways Won’t Scale

We all know the old ways to do the research, evaluation, and documentation:

Ask the person who’s been there the longest. In one job, we literally had a Q&A list called “Ask Jerry” to try to document key knowledge before one guy retired. This frankly does not work at scale. People are switching jobs faster than ever, the “retirement tsunami” continues to loom, and nobody devotes enough time (yet—cool tools are on the way) to externalizing their knowledge before they go.

Try to find some old versions and copy those. If you can’t find it, it may as well not exist. And if you only find the worst versions, you’re starting way behind in the race. Multiple times at the NRC, I was the one person on the project team with the perseverance to mine databases and microfiche and find key references that drove decisions. I’ve seen arguments won or lost based on finding a marked-up diagram from the early 80s. I’ve seen entire paradigms overturned because we found enough evidence that, yes, the decision-makers really did think about those valves before they decided.

Brute force it. We can call it “first principles thinking” and feel good about it (and sometimes that’s the right answer), but a lot of time it’s just reinventing the wheel because you don’t know the car’s already in the garage. There’s not a lot new under the nuclear sun. It’s a travesty that we spend people’s precious brainpower reinventing processes and knowledge that is locked in document treasure troves.

Everstar Brings a New Way

I’m here to lay out our approach. We see a gap in how this problem is being solved, and we want to close it. From others, you might get powerful search of your legacy documents, or rapid evaluation of the information in one system, or an AI buddy that can make your documents better, or even platforms that let you personalize once you learn how. But you really don’t want to fumble around in your tool chest while the toilet is leaking through the dining room ceiling. You want a plumber to come in and use all of his tools while you watch, and get the job done fast—as long as the price is right.

Our diverse team—to employ yet another analogy—is made up of foxes AND hedgehogs. We think about problems from many different domains. My deep and broad nuclear experience (new and old reactors, materials, policy). Founder energy and Silicon Valley speed. Intel and national security creativity. Years of technical design of hardware and software. Cutting-edge AI and groundbreaking search. This mix of experience gives us a range of understanding that lets us solve problems in totally different ways.

As a result, our tools tackle all aspects of that generalized view of the world—systematically and with amazing engineering rigor and quality. Here’s how:

If we don’t do ALL of these steps, we aren’t truly delivering the value that AI can bring. Poorly OCRed or organized information could lose data like a whole file cabinet going up in flames. Information that can be searched but not used is just a gee-whiz tool. Answers that still need a month of manual drafting to get regulator-ready aren’t really solving your problems.

That’s why every aspect of this chart is thoroughly researched, tested, and verified before it gets delivered to the customer. While of course we’re not going to give away the secret sauce in a blog post, here are some of the things that I’m most proud of:

OCR+++ that takes nuclear’s legacy documents (yellowed and crumbling as they may be) and turns them into information. When I saw illegible handwritten text on an early-80s deficiency report rendered into a structured table, I was blown away and made that the first thing I show customers.

Knowledge graphs—the under-the-hood tech that relates complex nuclear concepts across components, concepts, revisions, and jurisdictions. Our principal engineer built knowledge graphs for Microsoft way before those were cool (he’s got the patents to prove it). This sophistication is essential for reliable, rich results.

Nuclear-specific understanding that let you get the best results. Our tools understand the concepts so it’s the meaning that matters, not how someone typed. Do you want your opex searches to come up empty because someone might have typed AFW pump, aux feed pump, Terry turbine, or AFW-134B?

Collections of documents that allow focused gathering of insights. Why search the field when you know which haystack the needle is in? Our tech clusters together like documents (such as all measurement uncertainty uprate back-and-forth) and lets you chat with those collections. We know what you’re looking for. Why should you have to work hard to find it?

Highly complex evaluations that read like your best engineer wrote them. One of the first end-to-end modules we built guides 10 CFR 50.59 screenings and evaluations. It’s a complex process with tons of guidance and plenty of oversight gotchas. Our tool compares your situation to the criteria, tells you what it’s thinking and its confidence level, lets you add more information if needed, and (only when you’re happy) auto-generates the internal report, biennial report snippet, and FSAR markup. Twenty minutes. No BS. This is now my second-favorite thing to show customers.

Drafting that will unblock the industry. I’m sure you know how many thousands of hours are spent manually preparing documents and tables, usually pulling directly from your or someone else’s more detailed or earlier set of documents and tables. Then 5-10 layers of people review them for accuracy and noodle the words. Is this what your engineers went to school for? Is this what they like doing? We can take that burden away, generating long-form cited documents and tables in minutes, not months. You have to believe the inputs, or you’ll spend all your time micromanaging AI like your worst employee—so that’s why we engineer so carefully, trace our work, and cite our sources.

Jump on Board

This approach has resulted in tools that customers are calling the best in class. Some sneak previews, including a few things still in beta, are available on our solutions page. As we continue to listen to our customers and apply our unique philosophy, we’re confident we are going to transform this industry—and probably others.

If you resonate with what you read and want to make your life easier; if you are tired of editing and want to get back to engineering; if you’re in an adjacent industry and realize you have all the same problems—reach out. We are ready to nuke your paperwork, so you can focus on getting electrons on the grid.