Skip to content

Case study · hero · 2025 to present

Kintsu Medspa

Set the brand in the room with the client, then built the AI tooling that holds the bar after I leave the room.

Role
Hands-on designer, brand author, AI-systems builder
Company
Kintsu Medspa
Dates
2025 to present
Bucket
current venture
Live homepage of kintsuaesthetics.com showing a treatment room with sage walls, sheer curtains, and a white bouquet, with an editorial overlay panel reading Repair. Restore. Radiate. with Radiate set in italic, an eyebrow Physician and aesthetics, a Contact Us pill button, and an Explore Our Services link
Kintsu Medical Aesthetics · HomepageLive at kintsuaesthetics.com. Repair. Restore. Radiate. The editorial overlay sits against a treatment-room photo so the practice is anchored to the Fitzpatrick III to VI patient before the words start.

TL;DR

One-line outcome. I built the brand for a physician-led medspa in person with the client, encoded voice as a lint rule any AI contributor reads on every session, then shipped the site that runs against that system.

Role, dates, scope. Hands-on designer, brand author, and AI-systems builder for Kintsu Medical Aesthetics, a physician-led boutique medspa in Manassas Park, Virginia. Engagement began 2026-01-29. Repo initial commit 2026-03-24. Grand opening end of May 2026. I sit on the cap table. The website is the design and engineering deliverable inside the broader engagement.

Three metric chips.

  • 209 of 214 commits authored personally.
  • About 2,745 lines of brand and SEO strategy committed across six long-form documents.
  • SEO/AEO audit baseline 79 of 153 (52 percent) on 2026-04-06, with the matched site plan targeting 85 percent by Q1 2027.

Context

A physician-led medspa launching into a competitive local market, with a clinical specialty in skin of color (Fitzpatrick III through VI) and a primary demographic that requires Spanish-language reach. AI engines increasingly mediate first-touch discovery, so AEO presence is not optional.

The real problem was not the website. The bar for brand, voice, and AEO presence lived in my head and in client conversations, and the codebase was being built primarily by AI agents who needed to hold that bar on every session.

Mandate

Build the brand from a blank slate. Set the voice with the client face to face. Encode both as design files that govern every downstream decision (imagery, color, motion, IA, AEO, marketing funnel). Build the AI tooling that holds the bar. Ship the site: 17 pre-rendered routes covering home, about, eight service categories, pricing, FAQ, contact, careers, an opening landing, a thanks page, privacy, and terms.

Scope

Solo design and engineering ownership. The client partner is the Medical Director (Dr. Nawal, MD). No internal design or engineering team. I am the brand author, the systems builder, the reviewer, and the deployer. Pushes to main deploy the site to kintsuaesthetics.com via FTP to Hostinger.

Key decisions

Brand voice as a lint rule, not a tone-of-voice doc. I set the five voice pillars and the SAY THIS / NOT THIS pairs in person with the client. The cheap path was a PDF guideline and a hope that contributors would internalize it. I encoded voice as a prohibited-words table and a five-pillar contract inside repo-resident documents the AI reads on every session. The bar I set in the room is the same bar an AI hits at PR time.

Three-doc agent contract at the repo root. CLAUDE.md, AGENTS.md, and AI_DESIGN_GUIDELINES.md together govern every agent that opens the codebase. Each names exactly when to consult the brand story, the voice guidelines, and the marketing playbook.

Brand strategy as committed long-form documents, not slides. The brand story, voice guidelines, AI design manifest, marketing playbook, SEO/AEO audit, and site plan all ship as versioned files in the repo. The site reads from them. So does every contributor. Total: about 2,745 lines.

Operator-led AI tooling for strategic deliverables. The marketing playbook, the audit, and the site plan were produced by my own Manus skill (installed 2026-01-28). I briefed it, ran the Plan-Act-Observe-Update loop, reviewed every artifact, edited for voice, and committed it. The "Prepared by: Manus AI" byline is the generation signature of my own tooling, not a vendor handoff.

HIPAA-safe URL discipline encoded in the agent contract. CLAUDE.md and AGENTS.md ban patient identifiers from URL paths and query strings because GA4 (G-J41133T29R) captures page_location on every pageview. The rule lives where contributors read it, not in a separate compliance memo.

What changed

  • Brand: from blank slate to a kintsugi-anchored identity with a defended tagline ("Repair. Restore. Radiate.") and a clinical positioning that does not read like generic medspa filler.
  • Voice: from a writer's judgment call to a five-pillar contract with paired examples and a prohibited-words table that any reviewer (human or AI) can cite at PR time.
  • AI contribution surface: from "use AI carefully" to a three-doc contract that names which brand artifact to read for which task and bans the moves that would regress the voice.
  • Strategic deliverables: from consulting slides that get archived to repo-resident artifacts that the build reads, paired diagnostically (audit) and prescriptively (site plan).
  • AEO posture: from default-Google to a site that is explicitly machine-readable. robots.txt allowlists 13 AI bots including GPTBot, Google-Extended, PerplexityBot, ClaudeBot, anthropic-ai, Applebot-Extended, Bytespider, CCBot, and cohere-ai. llms.txt and llms-full.txt are generated and validated in CI.
The Kintsu Brand Identity page from kintsuaesthetics.com. Top ticker bar in deep teal scrolls clinical claims, then large serif Brand Identity heading, then the horizontal Kintsu Medical Aesthetics logo shown on a light cream surface and on a deep teal surface
Brand Identity page · kintsuaesthetics.comThe brand-identity page that lives on the public site. Same kintsugi logo, voice, and tokens that AI contributors read on every session, but readable in a single scroll by anyone the client hands the link to.

Measurable outcomes

KINTSU MEDSPA · 2026

Lines of brand strategy authored

Six committed long-form docs, 2026.

The bar lives in the repo, not in a slide deck.

Brand strategy committed as code, not handed off as decks.
  • 209 of 214 commits authored by Mohsin in the Kintsu-Site repo. Main branch alone is 152 commits as of 2026-05-10.
  • About 2,745 lines of authored or operator-led design strategy committed across six brand and SEO documents: Kintsu_Brand_Story_v3.md (308 lines), Kintsu_Voice_Imagery_Guidelines_v2.md (623), Kintsu_MedSpa_Digital_Marketing_Playbook.md (538), AI_DESIGN_GUIDELINES.md (71), seo-aeo-audit-report.md (450), seo-aeo-site-plan.md (755).
  • 17 public routes pre-rendered to static HTML via Puppeteer. Sitemap, llms.txt, and llms-full.txt regenerated and verified on every build.
  • SEO/AEO audit baseline 79 of 153 (52 percent) on 2026-04-06. Site plan targets 85 percent by Q1 2027.
  • Analytics live: GA4 G-J41133T29R, scroll depth at 25, 50, 75, 100, conversion event taxonomy (cta_click, email_click, phone_click, outbound_click, form_submit, form_engagement, view_service_category), HIPAA-safe URL discipline.
  • Grand opening confirmed end of May 2026.

Leadership lens

The same shape applies to a design org. Set the bar face to face with the team, encode it in tools they use every day, and the floor rises instead of the ceiling drifting. The strategic deliverables here (audit, site plan, marketing playbook) were produced by tooling I built and operated, not forwarded to a vendor.

What I did with my hands

Player-coach proof, all authored personally:

  • Kintsu_Brand_Story_v3.md (308 lines): kintsugi metaphor, "Repair. Restore. Radiate." tagline with each word defended in writing, clinical positioning, the Medical Director's voice quote.
  • Kintsu_Voice_Imagery_Guidelines_v2.md (623 lines): five voice pillars (Warm Not Soft, Knowledgeable Not Intimidating, Honest Not Salesy, Energetic Not Loud, Approachable Not Generic) with paired SAY THIS / NOT THIS examples, imagery direction, prohibited-words table. Authored off an in-person voice walkthrough with the client.
  • AI_DESIGN_GUIDELINES.md (71 lines): the brand condensed into directives an AI agent reads on every session. Kintsugi metaphor, voice pillars, color tokens with strict-adherence rules, motion easing, imagery rules, AEO directives.
  • The blocks-first composition system under client/src/components/blocks/ (PageHero, PageIntro, SectionHeader, FeatureColumns, LuxuryCard, CalloutCard, TreatmentTierGrid, ComingSoonWaitlist) so AI contributors compose from approved primitives instead of reimplementing patterns. A /blocks-preview route demos every block.

AI threading

Kintsu is a website build governed by AI contracts that point at the design files, plus a marketing layer produced by my own agent skill.

Three-doc agent contract. CLAUDE.md sets behavior for Claude Code (commands, brand color tokens, voice quick rules, HIPAA rule). AGENTS.md mirrors the contract for WARP and other agents. AI_DESIGN_GUIDELINES.md is the brand condensed for any agent that opens the codebase. AI labor was assigned by capability across design, debugging, and authoring tasks. For the full model-routing and skill architecture, see AI Fluency.

Operator-installed agent skill for strategic deliverables. The marketing playbook, the SEO/AEO audit, and the site plan are all outputs of my own Manus skill (installed 2026-01-28). Manus runs a Plan-Act-Observe-Update loop with citation tracking and a mandatory hallucination fact-check phase before delivery.

Machine-readability as a wedge. robots.txt allowlists 13 AI bots. llms.txt and llms-full.txt are generated from content.json at build time and validated in CI. Shipping llms-full.txt makes the content surface trivially scrapable. For a new practice fighting for AEO presence against established competitors, being machine-readable is the wedge.

Tech stack. React 19.2.1, Vite 8.0.7, TypeScript 5.6.3, Tailwind CSS 4.1.14, Framer Motion 12.23.22, Radix UI primitives. Testing: Vitest, Playwright across Chrome, Firefox, Safari, mobile Safari, Lighthouse CI. Pre-deploy gate: pnpm format:check && pnpm check && pnpm test && pnpm build:static && pnpm test:seo && pnpm test:links. Same gate runs in CI before FTP upload.

Reflection

The lesson is not "use AI to build a website." When the bar lives in the designer's head, the highest-value move is to set it with the client in person, then build the agent tooling and the repo contract that hold that bar after the meeting is over. Running this again as a Director, I would bring a design systems engineer in earlier to formalize the token layer, and pair the SEO/AEO audit with a quarterly UX writing review against the prohibited-words list to catch drift before it ships.