Work

Relay12

AI-powered field service support

Field service shops run on the knowledge inside a few senior people's heads. Relay12 lets junior techs text or call for answers, so the experts can stay on their own jobs.

Timeline

2025 - 2026

Live

Stack

  • Next.js
  • React
  • LLM + RAG
  • Voice / SMS
  • Tailwind
  • Vercel

Live

Visit live
Relay12 landing page - 'More jobs completed' headline beside a phone showing a tech troubleshooting a Carrier unit over SMS with cited manual sections
THE WHY

One stuck tech interrupts two people.

When a junior tech hits a problem on a job, they call the owner or a senior tech. Now two jobs are stopped instead of one. And the answer only lives in that phone call. Nothing gets written down for the next person who hits the same thing.

I kept hearing the same thing from trade businesses: the manuals exist, but nobody's reading a PDF on a roof.

WHAT I BUILT

Answers over text, from your own manuals.

Relay12 indexes a shop's own equipment docs and answers questions over SMS or a phone call, the tools techs already use in the field. No app to fumble with while wearing gloves.

Every answer carries a confidence score. When the model isn't sure, it says so and hands off to a person instead of guessing. In this line of work a confident wrong answer is the one that costs you, so I'd rather it stay quiet.

Relay12 conversation view - an SMS transcript of a tech asking for an equipment installation overview, answered step by step and scored at 80% confidence
A real exchange: an install walkthrough over SMS, confidence-scored before sending
WHAT MAKES IT DIFFERENT

Your docs, not the internet

Answers come from the shop's own manuals, so techs can trust what comes back.

It knows when to say "I'm not sure"

Low-confidence questions get routed to a person. The trust is the product.

Every answer sticks around

Each question and answer becomes a searchable record, so the shop's know-how stops walking out the door.

WHERE IT STANDS

Live, and in front of real shops.

The pitch that lands with owners is simple: your best people stop getting interrupted, and every answer becomes something the shop keeps. That second part is the bigger bet. The knowledge inside a small trade business is worth a lot, and nobody's really captured it before.

NEXT PROJECT

03

Parallax Mouse

Layered parallax for React

Read case study

Parallax Mouse demo - a 3D stack of labeled parallax layers shearing apart as the cursor moves, beside the package headline