SAGE — an AI receptionist for an NYC learning center.

Tracked, end-to-end. Built to convert.

I instrumented the site with full traffic and conversion tracking from day one — so every claim below comes with a number behind it.

What is SAGE?

SAGE New York is a multi-location academic enrichment and admissions-prep center for ambitious students across Queens and Long Island. Not a single-subject tutoring shop — it's a full pathway: K-12 tutoring, standardized test prep (SAT · ACT · AP · MCAT · LSAT), national academic competitions, research mentorship (SCI/SSCI publication), summer camps, and college consulting.

The brand serves families who plan ahead — parents who want structured support through multiple academic stages, from middle-school preparation to college admissions. The multi-location presence (Flushing, Little Neck, Jericho) anchors it in the NYC tutoring landscape where reputation and access matter as much as curriculum.

Two missing pieces: design discipline, and communication discipline.

The client had hired and lost three designers before me. Inbound traffic to the old site was decent — SEO was working — but conversion was bleeding out. Interested parents would land on the homepage and bounce. Each previous redesign attempt failed for one of two reasons.

Every piece of feedback is data. The tone is not the message.

I built a communication discipline before I drew a single screen. The MIT-grad liaison knew the business at a depth I never would — every sharp comment was a piece of information about what the business actually needed. My job was to separate the criticism from the delivery and work the criticism. When the tone got personal, I returned with a prototype. When the message was unclear, I asked rather than defended.

The tone softened the moment the system started compounding. By week 3 the liaison was forwarding my Looms internally — saying "this is how it should look." By week 8 we shipped on time, with the client referencing the project for their marketing.

Seven program tracks. Each one click from the homepage.

SAGE isn't a tutoring center. It's an ecosystem with seven distinct program tracks. The IA had to make that breadth scannable without burying any single track.

A seventh surface — Results — collects student outcomes (acceptances, score gains, competition placements) across all six tracks. Credibility lives there.

An AI receptionist that behaves like a teacher.

Most education sites have a contact form. SAGE needed something more — a receptionist that could answer parent questions 24/7 in their preferred language, classify their intent into one of seven program tracks, and route them to the right branch staff when the conversation needed a human. I named her Shirely.

The same pattern I used elsewhere applied here: LLM translates intent, deterministic code decides what's allowed. Shirely understands what a parent is asking. She doesn't decide whether their child gets a scholarship, which program to enroll in, or what it costs. Those decisions live in code and in human staff — never in the model.

Six tracks, plus an always-escalate.

Every parent inquiry is classified into one of seven closed categories. The taxonomy is fixed — not free-form — because parents asking the same thing in different words should land in the same place.

What Shirely won't do — and why that's the product.

Most AI chatbots fail by being too helpful. Shirely is designed with four boundaries that she'll never cross. The boundaries are constants — encoded in code, not prompts. The model cannot vote them away.

When in doubt, route to a human.

The model has confidence scores. Below 0.70, Shirely doesn't try to guess — she connects the parent with the appropriate branch director by name.

Three conversations. Three different exits.

How Shirely behaves in practice — same architecture, three different parents.

Brown, cream, and one accent. Education credibility, not startup gloss.

The design system leans on the values parents look for in an education brand: warmth, trust, longevity. Deep brown for primary surfaces (a SAGE brand color, evoking established institutions). Cream for content backgrounds. One red accent (SAGE's existing brand) reserved for proof points — accepted schools, score percentiles, key stats.

Typography is bold and decisive — parents scanning a page should know what each program does in 3 seconds. Body copy is plain English, not edutech jargon. Every page has a results section before its features section: proof first, claims second.

The site became the primary intake channel.

Before launch, the three NYC branches relied on phone-first intake — receptionists fielding calls during business hours, missed inquiries piling up after-hours. After launch, the site became the primary intake channel for all three branches, with Shirely covering the after-hours and multi-language gap. Branch staff spent more time on enrolled students; less time on first-touch screening.

The site also became the marketing anchor: every Google Ad campaign now points to a SAGE landing page, every print piece prints the URL, every flyer the branches hand out includes a QR code. Two months of design work; an ongoing return on attention.

Three things this project taught me.

I designed Shirely to behave like a teacher: patient, well-informed, never overselling, always pointing to the next step. The reason she works isn't that the model is good — it's that the boundaries around her are clear.