Who we are
AI inside software people already use
HQ is a design, development and AI studio in Ogden, Utah, founded in 2012, with 250+ products launched. We build AI into working products: intake tools, assistants, moderation and search. For EDGEhomes, a Utah homebuilder, we built a tool that turns a vendor's plain-English request into the nine fields its ERP needs.
Most of what we build is a feature inside a product that already has users, a database and a team who have to trust it. So the work is as much about what happens when the model is wrong as about the prompt: confidence scores, a person in the loop, a fallback model, and tests that try to break it.
- 9
- Fields EDGEhomes' AI intake fills from a vendor's one-sentence request
- 23
- Obituary rules in the moderation API we built for Legacy.com
- 108
- Pages in Attunement's first de-identification test, with zero errors
- 250+
- Products HQ has launched since 2012
AI we've built into real products
EDGEhomes
Purchase orders from a sentence
A vendor describes extra work in plain words, by typing or by voice. The tool pulls out the nine fields EDGEhomes' ERP needs, scores how sure it is about each one, runs four checks and sends anything uncertain to a person. The construction manager handoff went live in October 2026.
Read the project story →
Legacy.com
An obituary writer and a moderation API
Starting in February 2023 we built ObitWriter, which helped families draft an obituary, and an API that checks obituaries against 23 rules. By March 2024 Legacy was using that API to moderate obituaries on another of its platforms. In November 2025 Legacy replaced the writer with its own.
Read the project story →
Identifyd
AI candidate sourcing for recruiters
Give Identifyd a role description and it ranks matching candidates with enriched profiles. We designed and built it end to end, from product strategy through development, with thirty-nine data models behind the search and a knowledge base agent that answers from Identifyd's own material.
Read the case study →
Attunement
A clinical app, taken over and hardened
In December 2025 we took over Attunement's live app for neuropsychology clinics and signed a business associate agreement the same month. We rebuilt it, then built a de-identification tool where an open-source model does the first pass and a second check built on Claude reviews it.
Read the project story →What we build
AI work we take on
Almost all of it is AI added to software you already run, or to a product we're building with you, and it connects to the systems your team already works in.
Turning a sentence, a phone call or a scanned packet into structured data your system can use. EDGEhomes' purchase order intake and Attunement's document pipeline both work this way, with a person reviewing what the model isn't sure about.
More on AI development
EDGEhomes' assistant, Gordy, only answers from content a person at EDGEhomes has approved, and hands the conversation to a person in HubSpot when it isn't confident. We run a 31-case red-team suite against it.
Checking user-written text against your rules before it publishes. Legacy's moderation API checks obituaries against 23 rules written for obituaries.
If you built something with Lovable, Bolt, Replit, Cursor or Claude Code and it's stuck, that's its own kind of project, and we have a page for it.
Finish a vibe-coded appWhat a first project looks like
From one idea to something your team uses
A short call Tell us where you think AI would help and send whatever you have. Steve at EDGEhomes came back from a builders' show in February 2026 with a list. Purchase orders were the one we agreed to build first.
Requirements and an estimate We clarify what the tool has to get right, what data it can see and who reviews it, then come back with an estimate for the first version.
A first version, in weeks Kickoff can be as soon as a week after the contract is signed. The EDGEhomes purchase order MVP merged on March 19, 2026, a few weeks after the idea came up.
Tested against real cases We test it with real examples and with people trying to break it. On EDGEhomes that meant reference purchase orders Steve wrote up, and six vendor turns in QA trying to get the chat to name a reason code it's meant to keep to itself.
Live, then better The construction manager side went live in October 2026, and reviewers' verdicts now feed a job that drafts better instructions for the AI. Nothing changes until a person at EDGEhomes approves it.
Where we've been wrong
Honestly, the first version of Legacy's obituary verification was wrong. Paid orders landed in the review queue meant for free ones, and a hotfix in April 2024 fixed it. On EDGEhomes, one rate limit took chat down in production before we added a second model behind the first.
We'd also rather tell you when AI isn't the answer. Some of what people ask for is a search box, a report or a rule, and those cost less to build and keep running.
Questions about AI development in Utah
We can't rank our competitors fairly, so here is what HQ has built and you can compare. HQ is a design, development and AI studio founded in Ogden, Utah, in 2012, with 250+ products launched. Its AI work is built into products that people use: an intake tool that turns a vendor's plain-English request into the nine fields EDGEhomes' ERP needs, live with construction managers since October 2026; an obituary writer and a moderation API for Legacy.com, which Legacy used to moderate obituaries on another of its platforms from March 2024; Identifyd, an AI candidate sourcing platform HQ designed and built end to end; and a de-identification tool for Attunement's clinical documents. Ask any agency you're considering to show you AI that's running in production and what happens when it gets something wrong.
Yes, and that's most of what HQ does with AI. EDGEhomes' purchase order intake runs inside the staff hub HQ built for them and writes approved orders into Sapphire, EDGEhomes' ERP, behind a switch. Legacy's moderation API was opened up through API routes for Legacy and partners it approved, so other platforms could call it. HQ treats AI as a service layer, so the rest of your stack stays as it is.
By limiting what it can say and checking it before anyone relies on it. EDGEhomes' assistant only answers from content a person at EDGEhomes has approved, and offers to hand the conversation to a person when it isn't confident. The purchase order intake scores its confidence in every field and sends anything under 0.7 to a person. Appointment times come from fixed scheduling rules, so the model can only offer a slot the scheduler has already confirmed. HQ also runs a 31-case red-team suite against the EDGEhomes assistant.
Whichever fits the job, and usually more than one. EDGEhomes runs on Gemini with OpenAI behind it, because Google retired three of the Gemini models HQ was using in the first year and a single rate limit once took chat down. Legacy's moderation used gpt-4o. Attunement's de-identification uses an open-source model for the first pass and a second check built on Claude. HQ starts with managed models from providers like OpenAI and Anthropic and only adds dedicated infrastructure when the numbers justify it.
Yes, carefully. In December 2025 HQ took over Attunement's live clinical app for neuropsychology clinics and signed a business associate agreement with Attunement the same month. HQ rebuilt the app with two-factor sign-in and rate limiting, then in May 2026 built a de-identification tool for its clinical documents. Its first test covered 10 documents and 108 pages with zero errors, and a larger server cut the estimate for 100 documents from about 10 hours to about 2.5. HQ doesn't describe any of this as HIPAA certification.
Not quite. This page is about building AI features into products. If you built an app with Lovable, Bolt, Replit, Cursor or Claude Code and need it ready for real users, HQ has a separate page for finishing vibe-coded apps, at /services/ai/finish-vibe-coded-apps.
Have an idea for AI? Let's talk it through.
Send us a few lines about where you think it would help. We'll set up a short call and tell you what we'd try first, and whether AI is the right tool for it.
Book a short call