findaphone.ng

Challenge

Nigerians buying phones face opaque pricing, inconsistent specs, and outright scams across fragmented marketplaces. There was no trustworthy way to match a budget and a set of priorities to the right phone, or to tell a fair listing from a fake one.

Approach

Built everything in one Dart monorepo (Melos) so typed models and pure business logic travel cleanly across tiers: a Jaspr static site with @client interactive islands, a dart_frog REST API that loads the whole catalogue into memory at boot, a pure-Dart scraper, and Supabase Postgres with row-level security.

Curation Engine

As Library Curator, I vet every phone and listing that enters the canonical catalogue:

• Confidence-scored matcher (Jaccard similarity, brand multiplier, exact-model boost) tiers listings into auto-approve, review, and catalogue-gap buckets
• Admin review workflow protects already-reviewed rows from re-scrape clobbering
• Four idempotent enrichment pipelines (spec sheets, benchmark scores, review data, image rehosting), each writing its own column family
• Two-signal price-sanity filter catches listing scams, including a real mismatched-device listing in production

Evidence-Led Discovery

Extended the catalogue beyond specifications and prices with reviewed community OS support and an NCC registry browser. Official-source adapters produce candidates for human review, not automatic approval. Exact model variants, support status, and source evidence remain attached to published records.

• Community OS discovery published with reviewed support records across 566 phone families
• NCC registry browsing distinguishes exact model Type Approval from verification of an individual device
• Immutable Supabase evidence snapshots and repeatable publication tools preserve review decisions and provenance

Impact

Live in production on Cloudflare Pages, Railway, and Supabase with continuous deployment. Shared pure-Dart recommendation logic supports budget and preference-led discovery, guarded by unit, integration, schema-contract, and end-to-end tests. Recent work also covers server-component boundaries, session continuity, search indexability, and data-backed sitemap metadata.