Why I Built It

Some people who see Verdion in action ask, “…but who would use this? How does it help a business?” It’s an interesting question to me; I built Verdion because of a problem I saw in the way humans were relating to machines which I wanted to fix - I’m an engineer first, not a businessperson. That said, I know for a fact that Verdion can save businesses across a diverse number of verticals tremendous amounts of money.

So: The question these folks are asking, I think, is: “Who would use this, and why?”

Why They’ll Come: Use Cases

I think the strongest case for Verdion as it exists in MVP form, right now is made in any scenario where reasoning quality matters more than speed, and where the cost of a wrong answer exceeds the cost of Verdion.

Some Examples

• Contract review and legal analysis (see a real-world analysis)
• Medical differential diagnosis
• Systems and software architecture evaluation
• Code review and security auditing
• Strategic planning and decision-making
• Policy evaluation and risk assessment
• Research synthesis from conflicting sources
• Due diligence analysis

Overheard Remarks

When I showed the CableBurg Example Demo to a couple lawyer friends of mine, they instantly understood the value proposition: “I would have had to pay a team of paralegals for eight hours’ work to generate content like this.” To those of you who have never hired an attorney, I will point out that the cost of four paralegals working for eight hours far exceeds the cost of a Verdion case study - potentially by orders of magnitude.

Recursion as a Business Strategy

While building Verdion, I’ve noticed that many of the problems I’ve idly been wondering about with respect to my home lab/network, various small projects I’m working on, etc., could be prospects for Verdion - and I’m usually right.

The reason I bring that up is that there are a number of companies out there in the “multi-model AI” space who are effectively selling a deterministic routing algorithm: Their proprietary software knows that when you ask it for a code review, (e.g.) Opus is best at code reviews, so it gets routed to Opus. If you ask it a DB design question, it knows that (e.g.) Gemini is best-suited to those, so it routes the task to Gemini.

Verdion would be an excellent employee at such a company. Want to find out which model is best at a specific task? Queue up a few dozen/hundred/thousand of that task with known solutions against a Ring with your top 3 (or 4, or…) contenders. Statistically analyze the results. Now you know which model is best-suited to that task, and you have evidence to prove it.

Finally, I will mention again that the “eight-week plan” (I am taking it slower than eight weeks) I am following for Verdion is a plan authored by Verdion.

The point I am trying to make is: It seems Verdion is good at reasoning about reasoning. That kind of meta-cognition is rare in humans, and prized in C-Suites across the Fortune 500. The difference here is that Verdion doesn’t cost hundreds of thousands of dollars per year.

Post-MVP

I have a lot of ideas for Verdion. Some of them do involve that ugly word, execution - or orchestration. But Verdion’s flavor of orchestration would be innately heterogeneous: Not umpteen copies of the same LLM performing different tasks, but agents judged to be best at subtasks reasoned to be necessary to complete, managed by a series of neurons capable of canceling one another’s blind spots.

Beyond that, read the architecture document. The possibilities that become inherent in a multi-ring model where Rings and neurons are specialized for tasks of a certain nature are nearly infinite. And when Verdion can reason about how best to execute your task within its own framework, the potential only grows.