Skip to content
Adeshpreet Singh

4 featured builds · 7 shipped in total · average 13 days

Four builds, and the numbers behind them.

These are self-initiated concept builds, not client engagements. Each one starts from a constraint a real business would recognise, and ends with a measurement someone else can reproduce on a named profile.

Small Batch Brine interface
E-Commerce7 days

Small Batch Brine

Concept brief: a local food brand losing sales to Instagram DM chaos, needing a storefront without $39/mo Shopify overhead.

What I changed: Next.js storefront with a no-backend cart and a formatted WhatsApp receipt: orders land straight in the owner's inbox. Profiling found the layout preloading a 768 KB hero image the page never rendered: removing it took mobile LCP from 7.2s to 2.0s (median of 3 runs).

  • Order Intake

    Instagram DMs

    WhatsApp receipt

  • Platform Fees

    $39/mo

    $0/mo

  • Backend Required

    Shopify plan

    None

  • Lighthouse mobile · optimization (2026-10-15)

    74 · LCP 7.2s

    97 · LCP 2.0s

  • Lighthouse mobile · connection range (2026-10-15)

    150ms RTT: 99 · LCP 2.0s

    measured 16ms: 100 · LCP 0.7s

Read the build breakdownA commerce flow can be fast, fee-free and owner-editable without a backend.
MonthlyMuster interface
Web Application14 days

MonthlyMuster

Concept brief: a Kanban workspace where dragging cards feels instant and boards survive a refresh.

What I changed: Zustand state machine (isolated renders) + optimistic UI + Supabase Auth and Postgres persistence, with a one-click demo login so visitors skip signup.

  • Board Persistence

    Lost on refresh

    Zustand persist

  • Sign-in Friction

    Sign-up wall

    One-click demo

  • State Updates

    Full-board re-render

    Isolated store slices

  • Lighthouse mobile · connection range (2026-10-15)

    150ms RTT: 99 · LCP 2.0s

    measured 16ms: 100 · LCP 0.7s

Read the build breakdownComplex client state can survive a refresh without a full-page re-render.
TellersWindow interface
FinTech Dashboard21 days

TellersWindow

Concept brief: a fintech dashboard that stays usable on mobile: live prices, graceful fallback when the API fails, no blank screens.

What I changed: Type-safe TypeScript architecture with a resilient data layer: live CoinGecko prices backed by a seeded fallback so the dashboard never renders empty. Memoized Recharts components and aggregated series keep mobile scrolling smooth.

  • Market Data

    Static mock

    Live CoinGecko feed

  • API Failure UX

    Blank dashboard

    Seeded fallback

  • Type Safety

    Unchecked

    Strict, zero errors

  • Lighthouse mobile · connection range (2026-10-15)

    150ms RTT: 99 · LCP 2.0s

    measured 16ms: 100 · LCP 0.8s

Read the build breakdownA dashboard can stay useful when its API fails, and say so honestly.

Concept build: the figures shown are seeded demo data, not financial advice.

Marble Wood Brass interface
Agency Portfolio10 days

Marble Wood Brass

Concept brief: a luxury studio site that feels as bespoke as the interiors: cinematic scroll, editorial type, fast images.

What I changed: Next.js + GSAP for fluid scroll motion, content in typed local data modules, and next/image optimization (AVIF/WebP, long cache).

  • Images

    Unoptimized uploads

    AVIF/WebP, 30-day cache

  • Pages

    Template-locked

    Static prerender

  • Content Edits

    Developer-only

    Typed data files

  • Lighthouse mobile · connection range (2026-10-15)

    150ms RTT: 98 · LCP 2.2s

    measured 16ms: 100 · LCP 1.0s

Read the build breakdownContent-heavy pages can stay light when the images are treated as the design.

Named profiles, dated rows.

Lighthouse figures are not one number. Every row here is published with the profile and the date that produced it: Lighthouse’s own mobile preset (150 ms RTT, 1.6 Mbps, 4× CPU), the desktop preset where a build was measured on it, and, on most builds, a measured real connection (16 ms RTT, 4 Mbps). Where a build was optimized, the row it gained is a before and after on one profile, measured the same day. The gap between profiles is the finding: the connection assumption moves LCP by up to 3× on identical bytes. Any figure here can be reproduced on the same profile.

Every build breakdown

Want this applied to your site?

Start with the free teardown: the measured faults, the one fix worth doing first, and the five-day path. No price on the first note.