I don't usually do hackathons. I'm a designer who codes — I like things considered, deliberate, well-spaced. Hackathons are the opposite of that. Eight hours, two people, one shot.
But eMerge Americas 2026 had a Wildcard track for enterprise AI tools, and I had an idea I couldn't stop thinking about. So we went.
We placed 3rd. And we shipped something I'm actually proud of.
The problem we wanted to solve
If your company has remote employees in more than one state, you have a compliance problem — you probably just don't know it yet.
Every state runs its own labor law calendar. California updates its sick leave policy. New York changes its pay transparency requirements. Illinois rolls out new predictive scheduling rules. None of them tell your HR team. None of them sync to your employee handbook. And most companies are still tracking this manually — Google Alerts, legal newsletters, someone's bookmarks folder.
The average mid-market HR team spends 14 hours a month on compliance monitoring. 72% have zero automation. And US employers pay $8.3 billion a year in labor violations — most of them preventable.
That's the gap Canon is designed to close.
What Canon actually is
Canon is a Policy OS for HR teams. Not a chatbot. Not a compliance checklist. A full workflow tool that handles the entire lifecycle of a policy update — from the moment a regulation changes to the moment employees are informed.
The core loop looks like this:
- Monitor — Canon pulls live data from the Federal Register API and state DOL feeds. The moment a regulation is published, Canon sees it and filters it by your company's industry, states, and headcount.
- Analyze — The AI compares the new regulation against your existing policy docs and flags exactly what's non-compliant, down to the sentence.
- Update — HR edits via type, upload, or voice. Canon suggests compliant replacement text. The change routes to Legal for approval.
- Communicate — Once approved, Canon generates branded outputs automatically: an HTML announcement page, a Slack post, an email draft, and a PDF. All in your company's voice. All in seconds.
Every change is version-logged with a full audit trail — who edited, who approved, when, and which regulation triggered it.
The AI layer (this is the part I'm most proud of)
The thing that makes Canon different isn't the workflow — it's what the AI actually knows.
On setup, Canon ingests your company's existing policy handbook. Every new document you upload makes the model smarter about your specific language, tone, and coverage gaps. It's not working from generic HR law. It's working from your policies.
Then it connects to live government data. Not a weekly digest. Not a third-party aggregator. The actual Federal Register — the primary source — filtered in real time. When California publishes a new executive order at 9am, Canon knows by 9:01.
The comparison engine then does the hard part: it reads the regulation, reads your policy, and tells you exactly where the gap is. Not "this might be affected." Not "review section 4." The specific sentence that's out of date and a suggested replacement that's compliant.
We built this on the Anthropic Claude API. The reasoning quality is what made it possible to do this in 8 hours — the model can actually parse legal language and generate human-quality policy text without hallucinating nonsense.
What we built in 8 hours
- FastAPI backend with role-based access (HR, Legal, Compliance)
- Federal Register API integration with industry + state filtering
- Claude AI pipeline for gap analysis, policy matching, and text generation
- Three edit modes: inline type, document upload, and voice input
- Full approval chain with version control and audit log
- Branded output generator — HTML, Slack, email, PDF
- A landing page, a pitch deck, and a live demo
Two people. One day. It works.
The live demo is at uxmilena.github.io/eMergeHackathon and the pitch deck is here if you want to see how we presented it.
On the design
I wanted Canon to feel like something you command, not something you file paperwork into. The brief I gave myself: private jet UX. HR should feel like they're running compliance, not drowning in it.
The visual direction is editorial B&W with a fire red accent — #E8221A. Bebas Neue for display type, DM Sans for body. Thin linework as structural detail. No gradients, no cards, no dashboard bloat.
The pitch deck opens with:
HELLO ~~MOTHERF*CKERS~~ ⚠ STOP. That language is not compliant with your company communication policy. Section 4.2 — Professional Conduct Standards. — Canon AI, flagged 0.3s after input
It set the tone immediately. The judges got it.
What I learned
Building something real in 8 hours forces every decision to be intentional. There's no time to second-guess the design system, no time to refactor the component architecture, no time to bikeshed the copy. You make a call and you ship it.
The thing I keep coming back to: Canon isn't a hackathon project. It's a real problem with a real solution. The regulatory monitoring space is genuinely broken for mid-market HR teams, and the technology to fix it — live APIs, capable language models, structured approval workflows — all exists right now.
We built a proof of concept in 8 hours. A production version is a matter of scope, not possibility.
The team
Just the 4 of Us — which is funny because there are two of us. I handled product design, UX architecture, and the frontend. My teammate built the FastAPI backend, the Claude API integration, and the Federal Register pipeline.
3rd place at eMerge Americas 2026, Wildcard Track, Enterprise AI.
Not bad for a Saturday.
Canon was built at the eMerge Americas 2026 AI Hackathon on April 18, 2026 in Miami, FL. Live demo: uxmilena.github.io/eMergeHackathon Pitch Deck: https://uxmilena.github.io/eMergeHackathon/canon-pitch-deck.html Landing Page: https://uxmilena.github.io/eMergeHackathon/Canon.html