ennkar.com Platform

The problem

A reverse-mortgage lender needs a public site that educates homeowners, captures qualified leads, ranks for a genuinely competitive set of search terms, and hands staff a way to work those leads — without ever putting borrower PII where a content editor could accidentally expose it.

Constraints

• Content editors need CMS access without ever touching operational borrower or lead data.
• Lead attribution has to be trustworthy enough to report on, without overclaiming multi-touch or identity-resolution capability the system doesn't have.
• SEO output — ~150 location pages, comparison pages, guides — has to stay correct as content scales, not just at launch.
• Ad-platform conversion tracking needs to work under consent gating, without leaking PII into the conversion payload.

What I built

A Next.js 16 App Router site inside a pnpm/Turborepo monorepo, deliberately split across two databases: Payload CMS on Neon Postgres holds content only, and Supabase holds operational borrower and lead data. The stated rule is simple — no PII in the CMS — and the architecture enforces it rather than relying on discipline.

• A fail-closed lead-attribution pipeline: derivePersistedQuizLeadAttribution() computes attribution server-side in a Next.js proxy route, forwards it to a submit-quiz-lead Edge Function behind a shared secret, persists an allowlisted set of 10 metadata keys, and records 'unknown' for anything it can't verify rather than trusting client input.
• An OpenAI Ads pixel and Conversions API integration with a shared eventId for deduplication, gated on GPC and marketing consent, with no PII in the payload.
• A technical-SEO program: roughly 150 city location pages across 16 states, 146 of 150 with unique metadata, 6 competitor-comparison landing pages, and 30 learning-center guides, backed by a CI-integrated 'SEO autopilot' with a persistence gate, a circuit breaker, and auto-merge.
• Staff CRM surfaces with soft-delete, RLS, and audit triggers, capped Server Actions (a 200-ID limit), an open-redirect guard on return-to URLs, and a two-step confirm with a 6-second auto-disarm on destructive actions.

Architecture

Content and operational data are architecturally separated rather than merely access-controlled within one database: Payload on Neon for pages, learning-center content, and comparison landings; Supabase for leads, borrower interactions, and role approvals, with RLS on all ~26 public tables across 35 migrations. Roughly 11 Edge Function areas handle pricing, lead intake, Twilio Verify, FRED rate sweeps, role approval, and quiz intake — the same pricing Edge Function the mobile app calls.

Decisions and tradeoffs

Outcome

This is a publicly browsable platform with an unusually direct overlap with the requirements this portfolio itself was built against — technical SEO, structured data, and a CI-integrated regression program for exactly that.