Boutique AI studio · Portfolio

Selected projects, built where AI actually pays for itself.

Purple Kiwi AI builds working software for small and medium businesses — not strategy decks. Each project below started with a specific piece of operational friction: an audit that took an agency three weeks, a diagnostic nobody could run, a service bay quietly losing revenue. The model is never the point. The recovered hour, the closed deal, and the deliverable a client will pay for are.

Typical MVP sprint
2–6 wks
Fastest idea → live
24 hrs
Sectors
04
Owner per build
01

01Own product · Live

AuditSpark

auditspark.io

A website audit that scores a site for two audiences at once: the humans who land on it and the AI answer engines that decide whether to name it. Fifteen categories, a free GEO readiness check on every report, and a client-ready PDF with a 30/60/90-day plan.

Where AI does the work

Claude Opus runs the analytical judgment a senior brand-and-growth consultant would apply, across every category, on every page — while objective measurements (Core Web Vitals via PageSpeed Insights) stay outside the model so scoring can't drift. A second pass probes ChatGPT, Claude, Perplexity, Gemini and Copilot live and stores the verbatim answers as evidence.

The business value

A deliverable that used to take an agency two to three weeks of analyst time now lands in minutes, at a price a solo operator will actually pay. For agencies it inverts the economics entirely: a white-label report becomes a pitch asset and a retention artifact instead of a cost centre.

15Scored categories
6Engines probed live
minutesURL to full report
  • Claude Opus
  • PageSpeed Insights
  • PDF pipeline
  • REST API + MCP server
  • White-label tenancy
auditspark.io
AuditSpark marketing site

Report anatomy · six lenses

GEO
69
LLMO
46
AEO
34
SXO
6.5
SEO
6.0
E-E-A-T
7.5

Sample scoring pattern from a live AuditSpark report. Reports are AI-generated and advisory.

02Studio venture · Live

The Citation Lab

thecitationlab.com

The consulting-grade sibling to AuditSpark. Where the product scores a site, the Lab diagnoses a business across six layers and answers one commercial question: when a buyer asks an AI engine for a recommendation in your category, does your name come out of it?

Where AI does the work

The engines are both the subject and the instrument. Category questions are put to six answer engines on a weekly cadence, and the returned brand names, ordering and share of voice become the measurement — a live competitive index rather than a one-off opinion. Findings ship with the verbatim answer attached, so a client can check the work.

The business value

It turns an invisible loss into a line item. A crawler rule written years ago can be removing a business from AI answers entirely, with nothing in analytics to show for it. The free readiness check finds one concrete, verifiable fault before any money changes hands — a disciplined qualification step that makes the paid diagnostic an easy yes.

40%

Visibility gain, up to, in the foundational GEO benchmark — from evidence-led optimisation, not content volume.

17.3%

Citation-rate lift from structural change alone, content held constant, across six engines.

Crawlers per major AI company, each with a separate job. Block the wrong one and you vanish silently.

Published research results under specific conditions — cited by the Lab as evidence, never as guarantees.

  • Multi-engine probing
  • Crawler-access forensics
  • Schema architecture
  • Weekly re-measurement
thecitationlab.com
The Citation Lab website

The diagnostic stack

Diagnosed bottom-up — access problems make everything above them theoretical.

  1. L1Access — can AI crawlers reach you at all?
  2. L2Understandability — do engines know what you do?
  3. L3Trust — bios, credentials, real case evidence
  4. L4Citation-worthiness — is it structured to be quoted?
  5. L5Conversion — does the arriving visitor convert?
  6. L6AI visibility — are you named, and how often vs. rivals?

03Vertical tool · Pilot

Service No-Show Insights

servicenoshow.netlify.app

A no-show and capacity intelligence dashboard for dealership service departments, reading live appointment data from CDK Drive. Every dealer knows no-shows hurt. Almost none can say which advisor, which slot, or how much.

Where AI does the work

The arithmetic is deterministic — rates, lead times, revenue at risk — and deliberately kept that way. AI sits on top as the interpreter: it reads the completed cross-tabs and writes the paragraph a service manager would otherwise need an analyst for, naming the pattern, the likely cause, and the schedule change worth trying this week.

The business value

An empty bay is unrecoverable revenue with a known price. Attaching an average RO value to the no-show flag converts a vague operational annoyance into a dollar figure, ranked by day, slot and advisor — so the fix is a reminder cadence or an overbook rule, not a new hire.

  • CDK Drive · Fortellis API
  • Bulk + 48-hr delta sync
  • Server-side credentials
  • AI insight layer
Service No-Show Insights powered by CDK Drive
No-show rate
12.4%
Revenue at risk
$84.6k
Avg lead time
6.1d

No-show rate by day × time

AI-generated insight

Friday late-afternoon slots carry more than double the average no-show rate, and appointments booked over ten days out account for most of them. Tighten the reminder cadence for long-lead Friday bookings before adding capacity.

Rendered from the demo dataset shipped with the pilot; live deployments read a dealership's own CDK Drive records.

04Client work · Live

Lakou Wellness Collective

lakouwellness.com

Two halves of one engagement. Before launch: a waitlist that segments prospective clinicians by the kind of practice they want, fed by a lead magnet substantial enough to be worth an email address. From day one: the back office that actually runs the collective — scheduling, facilities, and client management.

Where AI does the work

Production leverage first — a 45-page workbook, intake forms, templates, planners and a prompt library drafted with AI, then clinically reviewed before anything shipped. Then in-product: a licensing assistant that walks a clinician through their state's requirements in plain language, instead of leaving them to parse a licensing board's website.

The business value

The collective opens without adding administrative headcount. Work that would ordinarily justify a coordinator hire — segmenting the waitlist, booking rooms, answering the same licensing question over and over — is absorbed by the platform on day one.

Back office

Scheduling, room and facilities booking, client management.

Licensing assistant

A guided bot walking a clinician through state licensing requirements in plain language.

  • Waitlist + segmentation
  • Scheduling + facilities
  • Client management
  • State licensing bot
  • Lead magnet system
  • Brand + content production
Lakou Wellness Collective

The Private Practice Launch Bundle

A clinician-friendly toolkit to launch, organise, and grow a private practice — six editable files unified into one system, delivered instantly on email capture.

  • Workbook
  • Social kit
  • Intake pack
  • AI prompts
  • Planners
  • Readme
45
Page workbook
6
Editable files
24mo
Founding rate lock
20%
First 3 months

Membership paths · waitlist segmentation

  • Virtual Professional presence Mostly virtual, wants a home base.
  • Hybrid Hybrid practice launch Newly licensed, building a caseload.
  • Dedicated Full-time office Ready for a private office.

05Consumer app · Live

BibleRacers

bibleracers.com

A family Bible trivia game with the friction removed: no account, no download, no setup. Tap the link and you are racing. Idea to live public URL took twenty-four hours.

Where AI does the work

The question bank is AI-generated and reviewed, which is the only reason a game like this is worth building alone — the content problem that would normally sink a trivia product becomes a weekend's work rather than a year of manual authoring.

The business value

Proof of the studio's actual claim, at consumer scale: twenty-four hours from idea to a live public URL. It is the shortest possible demonstration that a small idea does not need a quarter and a committee to reach real users.

0Accounts required
1Tap to first race
24hIdea to public URL
  • AI-generated question bank
  • No-auth session play
  • Mobile-first
  • Family-safe by design
bibleracers.com
BibleRacers — race through scripture, a family Bible trivia game

The through-line

Leverage, not theater.

Different sectors, different problems, and the same method underneath each one. The work starts with a workflow rather than a model, and finishes when something is live and measurable — not when the deck is approved.

01

Map the opportunity

Start with the workflow, not the model. What's slow, costly, repetitive, or hard to measure?

02

Design the product

Define the user, data, risks, success metrics, and the simplest experience that proves value.

03

Build the MVP

Ship the first working version in two to six weeks, with clean UX, secure architecture, and practical AI.

04

Launch & learn

Deploy, monitor, refine, and decide whether to scale, productize, or keep it internal.

Start small. Build something useful.

Have a workflow worth improving?

Send a note with the business problem, who's involved, and what success looks like. We'll help shape it into a practical first build.