Excellens.us
bilingual eyewear
retail SaaS.

The challenge.

Excellens needed a full e-commerce-grade retail platform — not a templated Shopify swap, but a custom build that matched a bilingual customer base, the operational realities of an optical shop (frames, prescriptions, fittings), and an admin back office the in-house team could actually run.

The constraints that ruled out the off-the-shelf options:

• Bilingual from the URL up. Every public page in English and Spanish, language toggle preserved across navigation, all product copy localized — not just a translation widget bolted to the top right.
• Operationally real. Appointment booking with the right service-to-staff mapping, frame inventory that reflects the actual stock on the shelf, an admin role with JWT-protected endpoints, and an editor experience that doesn't require a developer to update a price.
• Production-grade from day one. Sticky-nav-and-Radix-dialog UX, server-side rendering for SEO on product detail pages, image hosting that scaled past the seed-data phase, and an admin dashboard the owner could log into without a tutorial.

The approach.

PickBits builds with AI from the first commit — but the discipline behind it is regular, boring, senior engineering. The Excellens build followed the same three-phase pipeline taught in the Vibe Coder track: Explore, Validate, Harden.

1. Explore — the 30-minute prototype

First, a working prototype with the hero, the product grid, the appointment flow, and a single-language-only catalog. Real fonts, real photographs, real product data scraped from the existing Mutual Optical source the client already owned. The goal wasn't perfection — it was "Excellens, does this feel like your shop." Two iterations against the founder's reactions and the shape of the build was locked.

2. Validate — real inputs, real edges, real users

Then: the bilingual layer, the admin back office, the JWT auth flow, the product detail pages, the appointment modal. Every flow tested against the messiest possible input — incomplete frame data, customers who switch language mid-booking, staff editing inventory from two browser tabs. The transition out of Validate wasn't "it works on the happy path" — it was "the edges don't ship surprises."

3. Harden — boring discipline before launch

Finally: SEO meta tags, Google Analytics wiring, Radix-dialog focus-trap fixes, sticky-nav scroll behaviors, the framer-motion entry animations that don't tank Lighthouse, the seeded PostgreSQL frame catalog migrated from JSON, the admin auth flow stress-tested, the production deploy on the operator's domain. A stranger can run it now without paging anyone.

The stack.

Every choice serves a real constraint, not a resume bullet:

Tailwind for the velocity (no design-system overhead for a focused build), React for the team's existing fluency, PostgreSQL because the frame catalog is genuinely relational (frames belong to brands, have variant colors, link to vendor sources). Radix UI for the appointment modal because accessibility was non-negotiable. JWT for admin because the back office is the only thing behind auth — no over-engineered identity provider for a small operation.

The results.

Live, indexed, and operating. Real customers booking real appointments through the modal flow. The admin role manages frames without a developer in the loop. Bilingual customers don't bounce on the language toggle.

What the work looks like.

The same operator-led delivery is available for your build. If you have a real business with real customers and an idea you're tired of describing on calls, scope it on a free 30-minute call — engineers, not slides.