All work

Platform Liberation

Five buyback brands. One codebase. Fair offers enforced at the system level.

Most electronics buyback platforms fail in one of two ways: quote high and deduct aggressively at receipt, or quote conservatively and lose to competitors. We built a platform where fair pricing is a system constraint, not a policy — and scaled it to five brands from a single Next.js codebase.

Next.jsSupabaseMulti-tenantVercelAI Quote EngineEasyPostResend
5
Brands
one codebase, one database
60 min
Max price lag
upstream sync via Supabase cron
$0
Proxy pricing
hard rule — unknown model = decline

The Problem

Quote high online, deduct at receipt. That one pattern kills more buyback businesses than anything else.

A vertically integrated electronics reseller needed five customer-facing buyback brands targeting different segments — general laptops, MacBooks, gaming laptops, local repair market, iPhones — each with its own SEO identity, pricing rules, and email branding. Building five separate codebases was off the table. The pricing had to be accurate: quote high and deduct later is the fastest way to destroy trust in this category.

What Was Built

Multi-Site Architecture

Next.js 14 + Edge Middleware
  • Five domains share one Next.js monorepo — middleware resolves incoming domain to site_id at the edge before any route handler runs
  • Branding, email routing, category filtering, pricing display all driven by site_id — no duplicated code
  • Adding a sixth site is a database and DNS operation — core logic unchanged
  • Per-site category filtering: MacBook site shows only Apple devices, iPhone site shows only phones and tablets

Pricing Engine

Supabase + Hourly Cron
  • All prices sync from upstream wholesale database via Supabase-to-Supabase connection — no API polling, no rate limits, no stale cache
  • $0 upstream price = hard DO NOT BUY signal — no fallback pricing, no proxy models, no exceptions
  • Three-layer pricing: base price × condition multiplier − defect deductions, all category-aware
  • Deduction percentages are admin-only — customers see the dollar offer, not the math behind it

Sara AI — Quote Assistant

Claude Haiku + Extraction Pipeline
  • Customer types natural language — Sara extracts brand, model, RAM, storage via Haiku before hitting the database
  • Extraction step eliminated the entire class of failures from noisy customer language polluting search queries
  • Exact model match with non-zero price: Sara quotes and links to full quote flow. No match: routes to brand/category page — never dead-ends
  • Order lookup by order number in-chat — customers check status without leaving conversation

Operational Intelligence + Retention

Resend + Supabase + EasyPost
  • Every new order triggers admin email: payout, upstream wholesale price, estimated margin, eBay sold comps, parts candidate flag
  • Quote expiry at 14 days with automated sequence: reminder day 3, follow-up day 7, last-chance day 10
  • Win-back cron: expired orders re-engaged with current price — if price increased, customer is told explicitly
  • $20 shipping minimum enforced at library level — runs before EasyPost is called, cannot be bypassed by alternate entry points

Search Performance

LIVE

Pulled live from Google Search Console API · Jul 31, 2026, 8:14 PM UTC

2K
Clicks
last 28 days
89K
Impressions
last 28 days
4K
Clicks
last 90 days
249K
Impressions
last 90 days

Top Pages (90 days)

/
822clicks
27.1pos
/laptop
279clicks
44.1pos
/sell-broken-laptop
191clicks
9.1pos
/laptop/hp
166clicks
20.9pos
/laptop/lenovo
126clicks
11.6pos

Top Queries (90 days)

sell my laptop
170clicks
16.2pos
sellmylaptops
108clicks
1.0pos
sellmylaptop
74clicks
2.7pos
sell laptop
30clicks
24.2pos
sell laptop online
28clicks
29.0pos

Key Lessons

01

The $0 rule is non-negotiable.

One instance of proxy pricing — quoting based on a similar model because the exact model had a $0 price — resulted in a cancelled order and a customer apology. The rule was hardened: $0 means decline, always.

02

Extraction before search.

Raw customer language is noisy. A lightweight AI extraction step between customer input and database query dramatically improved match rates without meaningful cost increase.

03

Safety checks belong in the library, not the route.

The $20 minimum was in the standard label function but missing from the bulk order route. Safety checks live at the lowest possible layer so they cannot be bypassed by alternate entry points.

Ready?

Build your own buyback platform.

Start a Project