Built by a lawyer who got tired of citations that lie.

I'm Matt Kading. For decades, I've been the lawyer in the room insisting we argue about what the law actually says — in courtrooms, through marathon briefing sessions, and in the occasional regulatory meeting where everyone politely pretends to understand blockchain.

I've drafted motions on zero sleep, agonized over footnotes at 3 a.m., and once listened to opposing counsel cite authority so spectacularly wrong I briefly wondered if they'd been replaced by an early version of a chatbot.

That experience led me to become deeply obsessed with building AI for legal research, writing, and litigation prep — but only the kind that can actually prove it read the sources.

Lexora is the result.

It retrieves real authorities, shows its full reasoning trace (like a diligent associate who knows you'll check every citation), quotes verbatim, and has the professional integrity to admit when it found nothing useful. No hallucinations. No fabricated cases. Just honest, verifiable work.

In an industry flooded with legal AI that overpromises and underdelivers, Lexora offers something refreshingly rare: accuracy you can trust — delivered with clarity and transparency.

Because in law, confidence without proof isn't intelligence. It's malpractice waiting to happen.

Welcome to a better way to practice law. Welcome to Lexora.

Why retrieval-first (aka "stop making up cases")

Generic legal AI has one superpower: sounding like a fifth-year who drank too much coffee and not enough Bluebook. The tone? Impeccable. The citations? Fiction. I've seen tools produce holdings so polished you almost forget to ask the only question that matters: did you actually pull this?

Lexora's rule is stupidly simple and weirdly controversial: cite only what you retrieved in this session. Show the queries. Show what got excluded. Flag the gaps. Run a verification pass. It's less "magic legal oracle" and more "research with receipts" — which, if you've ever had a judge raise an eyebrow at a citation, you'll agree is preferable to magic.

Used properly, AI can shrink research from "cancel your weekend" to "still have dinner." It can help you draft faster without turning your brief into a beautiful lie. It can prep you for litigation the way a great paralegal does — except this one doesn't call in sick during trial week. The catch: the system has to be grounded. Otherwise you're just speed-running malpractice with better typography.

My other AI habit (it's fine, I can stop anytime)

When I'm not torturing language models with Erie doctrine, I train them to play tower defense games. I'm not kidding. Choke points. Upgrade paths. The agonizing question of whether this wave needs splash damage or whether you're emotionally attached to a tower that hasn't gotten a kill since 2019. I have strong opinions about creep leakage and stronger opinions about reward functions.

Is teaching a neural net to survive wave 47 the same skill as legal strategy? Obviously not. But both involve optimizing under constraints, punishing bad decisions, and watching something you built get wrecked because you skipped a step. In law the step is "verify your cite." In tower defense it's "maybe don't sell your only ice tower before the flying units." Life lessons.

Community AI (the fun part)

I also build small local AI apps for people around me — tools that run on their machines, don't harvest their data, and don't charge $49/month to answer "how do I fix this form." Helping someone untangle a bureaucratic knot, draft a letter a human will actually read, or finally understand a problem they've been stuck on since last tax season — that's the stuff that reminds you why you learned to code in the first place. Well, that and spite.

What Lexora is (and isn't)

Lexora is not your replacement. It's not going to argue the motion for you, charm the jury, or know that Judge So-and-so hates adverbs. It will run a real research loop, stream a trace you can actually audit, and give you citations you can defend without whispering a prayer under your breath.

If that sounds useful — try the app, skim the blog, or say hi. I'm always happy to talk retrieval, bad citations, optimal tower placement, or why "the AI said so" is not a litigation strategy.