December 2023. A prototype I built as the data layer for BorderDollar — a trade finance startup for maritime SMEs.
the problem
Credit risk in maritime trade is hard to price. Counterparties span multiple jurisdictions, geopolitical shocks move fast, and most of the signal lives in industry news — not structured data. BorderDollar needed a way to continuously monitor maritime and shipping news to surface risk signals for a future NLP scoring pipeline.
what I built
MediaDollar is a three-part system: a scraper, an API, and a dashboard.
MediaDollarDB — a Python script running on a 6-hour GitHub Actions cron that pulls RSS feeds from six maritime publications:
- Splash247
- The Maritime Executive
- Ship & Bunker
- Manifold Times
- FreightWaves
- Hellenic Shipping News Worldwide
Each article is parsed, deduplicated by title, and written to a PostgreSQL database.
MediaDollarAPI — a FastAPI backend deployed on Vercel, exposing a paginated /latest endpoint returning articles sorted by recency.
MediaDollarApp — a React dashboard (Chakra UI) showing a consolidated feed of all articles with infinite scroll.
why
The longer-term idea behind BorderDollar was to process this feed through an NLP pipeline — tagging articles by risk type, scoring counterparty exposure, and feeding signals into a credit scoring model for maritime trade finance. MediaDollar was the data layer: get the collection pipeline running reliably before building the intelligence on top.
stack
Python, feedparser, PostgreSQL, GitHub Actions — RSS scraper and DB layer. FastAPI, Vercel — API. React, Chakra UI, Vercel — frontend.
what happened
The scraper ran reliably on its 6-hour cron and the dashboard consolidated news across all six sources. BorderDollar eventually shut down in December 2024 after not hitting PMF — MediaDollar was the data infrastructure we built while working toward the NLP scoring layer that never got finished.