We launched 7 B2B APIs on Product Hunt — here's the full stack

#showdev #producthunt #api #cloudflarechallenge

We just launched JSONexLab on Product Hunt — a suite of 7 B2B APIs for trade, finance, and compliance. Free tier on everything.

👉 JSONexLab on Product Hunt — would love your feedback!

What we built

After ~9 days of aggressive sprint, here's the full set:

HS Codes Classifier — AI classification for customs (75.5% HS2 accuracy, beats GPT-4o-mini baseline)
Compliance Engine — sanctions screening (OFAC / UK / EU / UN + 11.7K PEP + 20K GLEIF entities, 33K+ sanctions records)
Docs-to-JSON — document extraction (invoices, receipts, contracts) in ~2 seconds
Finance Validators — IBAN / VAT / SWIFT / LEI / Card validation
Watch Engine — currency rates, VAT rates, tariff rates (203 monitors)
Creator Engine — promise extraction from contracts, NDAs, legal docs
ScopeGuard — sanctions scope monitoring with webhook alerts

Plus 5 MCP servers on the official Model Context Protocol Registry:

document-to-json-mcp
watch-engine-mcp
compliance-engine-mcp
creator-engine-mcp
scopeguard-mcp (v1.0.1, published today)

The stack

Cloudflare Workers — every API runs at the edge
D1 (SQLite) — for compliance records
KV — for caching
R2 — raw data storage (OFAC XML ~29 MB, UK OFSI CSV)
Pages — static frontend
Git-driven deploy — git push → live in 30 sec
x402 payments — USDC on Base ($0.01–$0.05 per call)

Architecture notes

Sanctions ingestion pipeline (GitHub Actions → Worker):

GitHub Actions cron runs daily at 01:00 UTC
Fetch OFAC XML from sanctionslistservice.ofac.treas.gov
Upload raw XML to R2 (compliance-raw bucket)
POST /cron/parse/ofac on Cloudflare Worker
Worker reads R2 in chunks
Parse XML → normalize names → UPSERT into D1
/health/ingestion endpoint reflects status

Why GitHub Actions for fetch? Cloudflare Workers get HTTP 525 from Treasury's Akamai-fronted endpoints (SSL handshake fails between CF IP range and Treasury). GitHub runners don't have this issue. So fetch lives in GitHub, parse lives in Worker.

Name normalization: NFD Unicode decomposition → strip diacritics → transliterate Cyrillic → strip legal suffixes (LLC, GmbH, OJSC, etc.) → lowercase tokens. Boosts fuzzy match precision without ML.