Corporate Intelligence Platform

Map corporate ownership across 65,000 companies for consumers making boycott decisions.

What was built

React + Vite frontend on Cloudflare Pages, with iOS app and Chrome extension as additional surfaces.
Supabase Postgres (with row-level security) for users, accounts, and boycott data.
Neo4j AuraDB for a 65K-company / 38K-brand / 77K-product knowledge graph; subsidiary, supplier, ownership, brand-of relationships modeled as typed edges.
Express API on AWS App Runner serving Cypher queries, fronted by a four-tier read cache.
Daily Python ETL on AWS Lambda enriches the graph from public registries; idempotent, resumable, checkpointed across 27 stages.
21K company logos batch-uploaded to R2, served via Cloudflare's CDN.
Sentry APM tracing across the API; rate-limiting, sanitized error messages, security audit completed.

Stack

React 19, Vite, Tailwind CSS 4, Supabase (Postgres + RLS), Neo4j AuraDB, AWS App Runner, AWS Lambda (Python ETL), Cloudflare Pages, Cloudflare R2, Cloudflare Workers KV

Outcome

P95 company-detail latency ≈40ms via four-tier caching: edge cache → Workers KV → in-memory LRU → R2 static fallback.

What was hard

Reconciling conflicting corporate-ownership data across five public registries (GLEIF, Wikidata, SEC EDGAR, FEC, OpenCorporates), each with different ID schemes and naming conventions, and keeping every relationship stamped with source + confidence + last-confirmed metadata so a journalist could audit the citation chain.