Brux ships a dental AI SaaS with 4 capture channels and end-to-end GHL sync

The situation

Jake Efstration, founder of Brux Dental Solutions, saw where dental marketing actually leaks. Most practices do not have a lead-generation problem. They have a conversion capture problem: a patient gets curious about cosmetic work, cannot picture the result, says "let me think about it," and never books. Show them their own face with a new smile, and the conversation changes. And no smile preview tool on the market plugged straight into GoHighLevel, the CRM his dental office customers already ran on.

He had a prototype of the idea: a no-code AI smile tool that demoed well. But it could not be sold. It had no billing, no way to run many offices on one platform, and its output was too inconsistent to put in front of real patients. He asked us to turn it into a real product.

We built Brux Smile Makeover, a dental marketing SaaS where a patient uploads a selfie and gets a lifelike new-smile image in 30-40 seconds, plus an optional before/after reveal video. Patient photos come in through four channels — a public webform, in-office upload by staff, an external webhook, and photos texted to the office's GoHighLevel number — and all four run through the same AI pipeline. Results flow back into GoHighLevel on their own: the patient becomes a contact, treatment tags are applied, and a text with their new smile lands on their phone.

Phase 1 went live in April 2026. In June 2026 we added self-serve office sign-up from printed QR flyers and an Enterprise tier where offices bring their own AI key. A third phase is under discussion.

What we had to solve

• Getting an unpredictable AI model to produce natural-looking smiles every time

  The whole product depends on one question: does the new smile look real? Instant selfie apps get treated as filters, and a preview that looks fake loses the patient's trust instead of winning a booking. Image models make this hard, because the same prompt gives a different result every time. The quick route was to take a fast model and ship it. We pushed back, because fast output that looks wrong would sink the product. Instead, we compared several image models and AI providers, reworked the prompts until "change only the teeth, never the face" held up on real patient photos, and added a check that rejects unusable photos before any credit is spent. With that groundwork in place, the fast model became good enough. The platform runs on Gemini 2.5 Flash Image at 30-40 seconds and about $0.034 per transformation, and each patient gets three results to choose from.

• A credit system that stays correct when a busy office submits many photos at once

  Every AI render costs real money, so offices pay in credits, and a job has to stop the moment an office can't cover it. That check runs across four entry points, on AI jobs that finish in the background and can fail halfway. Our first version used a standard optimistic lock, and it broke on busy offices: when two submissions landed at the same moment, one was rejected even though the office had plenty of credits. For a clinic running several previews during a consult, that's a patient turned away for no reason. We replaced it with a single database update that reserves credits only if the office stays within its limit. Now every submission goes through while credits remain, and only a job that would actually go over the limit is stopped. Credits are charged only when a result comes back. If a job fails, its reserved credits are released.

• Running the entire AI pipeline on the dental office's own API key

  Enterprise offices run every AI call on their own OpenRouter key instead of Brux's. That meant finding every step that quietly used Brux's key: photo validation, the image variants, the video start frame, and the video itself. One function now decides, for each submission, whose key it runs on, and every later step follows that decision. If the office's key runs out of budget, the platform says so clearly and never falls back to charging Brux. One surprise along the way: OpenRouter returns a 401 error, which normally means "invalid key", when a key hits its spending cap. At first, admins were told their key was broken when it only needed a top-up. We now read the error details to tell the two apart.

1. AI turns a patient selfie into a lifelike new smile in 30-40 seconds, with three results to choose from

   A patient uploads a front-facing photo. Before any credit is spent, Gemini 2.5 Flash checks the photo is usable. Photos that are not of a person, cropped too tight, too blurry, or show more than one face are rejected with a clear reason. If the photo passes, the pipeline generates a base result and variants with stronger correction in parallel, using Gemini 2.5 Flash Image. By default, a webform submission shows three results side by side. The patient picks a favorite and can ask for a before/after reveal video: a second AI pass creates a closed-mouth starting frame, then Veo 3.1 animates the smile opening.

   

2. Patient results land in GoHighLevel as contacts with tags, custom fields, and a text message, automatically

   Jake saw GoHighLevel integration as the product's edge in the market. When a transformation finishes, the platform finds or creates the patient's GHL contact by phone number, applies treatment tags, saves the before/after image and video links into GHL custom fields, and texts the new smile to the patient. It works in both directions: when a patient texts a photo to the office's GHL number, the platform picks it up, runs the full pipeline, and texts the result back. Every submission, whichever channel it came from, links to its GHL contact, so staff can jump to it straight from the Brux dashboard.

   

3. Offices claim their account and start a trial from a printed QR flyer, with no setup from Brux staff

   Brux finds new offices with printed flyers, mailed to each practice with a unique QR code and a 9-digit claim code. When a dentist scans the code, they verify their email and phone number and activate a trial account that already has their webforms set up. Once the flyer is in the mail, Brux staff don't need to do anything. Claim codes are stored hashed for lookup and encrypted for admin display. Activation runs as a two-step database transaction with row locks, so the same office can never be claimed twice. If activation fails, the login account created for it is rolled back.

   

4. Full Stripe billing with credits, trials, coupons, auto-reload, and an Enterprise tier that skips credits entirely

   Offices subscribe to Basic ($99/mo, 100 credits), Advanced ($199/mo, 300 credits), or Enterprise ($299/mo, no credits). Each generated image costs 1 credit, so a three-result submission costs 3, and a video costs 4, charged only when the patient asks for one. The card is saved during the trial and charged automatically when the trial ends. The Enterprise tier skips credits entirely and runs AI on the office's own OpenRouter key. Plan changes, coupons, and 3-D Secure checks on auto-reload payments all run through Stripe, with a webhook handler keeping each office's plan, credits, and coupon balance in sync.

   

How it runs

React 19 + Vite

We built the UI from the client's Figma Make designs, and feedback came in every week. React 19 with Vite let us turn each design into production screens quickly and ship UI changes without slowing the backend work.

AWS Amplify + Lambda

Amplify let us define a Lambda function for each area (billing, AI transforms, GoHighLevel, webhooks, auth, scheduled jobs) without running an application server. Each transformation runs as its own Lambda call, so a rush of submissions from one office doesn't slow down another.

Hasura + PostgreSQL

Hasura event triggers drive the AI pipeline: each new photo upload starts the transform job, which reserves credits and calls the AI models. We generate a typed SDK from the GraphQL schema, so every Lambda reads and writes data the same type-safe way. The credit limit check is a single conditional database update, so it stays correct under concurrent load.

Google Gemini and Veo

Gemini 2.5 Flash Image produces smiles good enough for a dental brand, in 30-40 seconds at about $0.034 per transformation. It also creates the closed-mouth start frame for videos, while Gemini 2.5 Flash checks each photo first. Veo 3.1 creates the before/after reveal video. Enterprise offices run the same models through OpenRouter on their own key, and can pick Gemini 3 Pro Image or Veo 3.1 Fast to trade speed for detail. We compared several models and providers, and this combination met the quality bar while keeping speed and cost within the product's budget.

Stripe

Stripe Subscription Schedules let offices downgrade without losing access mid-period. A SetupIntent saves the card during the trial and charges it when the trial ends. Coupons live in Stripe, with a credit balance per office that renews, expires, and follows the office through plan changes. Trial conversions, coupon switches, 3-D Secure on auto-reload, and an Enterprise tier with zero credits all run on Stripe's billing tools, so we only wrote the logic specific to Brux's credits.