WG7 Empreendimentos

Highlighted implementation

One slice
that tells the story.

// List properties from URL searchParams — type-safe + paginated
type Filters = z.infer<typeof PropertyFiltersSchema>;

export async function listProperties(
  filters: Filters,
  page = 1,
): Promise<{ items: Property[]; total: number; pages: number }> {
  const supabase = createServerClient();
  const PAGE_SIZE = 24;

  let query = supabase
    .from("properties")
    .select("id, title, slug, area_m2, monthly_rent, photo_url, city, state, category, featured", { count: "exact" })
    .eq("status", "active");

  if (filters.state) query = query.eq("state", filters.state);
  if (filters.category) query = query.eq("category", filters.category);
  if (filters.listing) query = query.eq("listing_type", filters.listing);
  if (filters.minArea) query = query.gte("area_m2", filters.minArea);
  if (filters.maxRent) query = query.lte("monthly_rent", filters.maxRent);

  const { data, error, count } = await query
    .order("featured", { ascending: false })
    .order("created_at", { ascending: false })
    .range((page - 1) * PAGE_SIZE, page * PAGE_SIZE - 1);

  if (error) throw new ListPropertiesError(error.message);

  return {
    items: data ?? [],
    total: count ?? 0,
    pages: Math.ceil((count ?? 0) / PAGE_SIZE),
  };
}

Retrospective

What I'd do
differently today.

Decisions I'd reverse knowing what I know now. Not regret — the kind of learning that changes the next project.

1. 01

   Structured lead tracker from day one

   Leads go 95% to WhatsApp and live in chat history. Functional for this phase, but loses valuable funnel insight — which city converts best, which category attracts leads but doesn't close, which button copy performs better. The leads table exists, but with too few fields. Next iteration: UTM + origin (applied filter) + minimal status (new, in conversation, won, lost).

2. 02

   Specialized image CDN before scaling the portfolio

   Supabase Storage handles the current volume (15+ cities, ~100 properties), but optimization is on-demand via next/image — first request per new photo pays 800ms. When the portfolio hits 500+ properties, Cloudflare Images or Bunny CDN with upload-time transforms become the right call. It's a timing decision, not an architectural one.

3. 03

   Single parametrized Map component from the first use

   Leaflet was added case-by-case as new places asked for a map. Today there are 3 spots with slightly different configurations. Should have been one parametrized Map component from the first instance — a refactor sprint will fix it, but the cost was zero if done directly.

4. 04

   A/B testing on the main CTA copy

   The contact button is fixed: 'Talk to WG7'. For corporate B2B, copy matters: 'Schedule a visit', 'Negotiate lease', 'Request a quote'. I'll run a simple test with a per-city flag — measure click rate without investing in heavy tooling. Lack of discipline measuring this early on.