AI Content & Geolocation Pipeline

NewsMap is a local news and events app I founded and have been building full-time since September 2024. At its core is an AI content pipeline that aggregates and geolocates local news, turning scattered, unstructured articles into a reliable, personalized feed.

When I first posted NewsMap as a side project on Reddit, it got far more attention than I expected: 77K+ views, 768 upvotes, and 169 comments, with 99% positive sentiment.

The pipeline runs in three stages, each one narrowing the set of articles and adding cost and precision. Ingestion pulls from RSS feeds and scraping, with fallbacks for sites that actively resist bots. A cheap AI pre-filter then asks a simple question of just the headline: is this even about a place? Only the survivors move to the expensive stage, where AI extracts a candidate location, Google's Places API geocodes it, AI scores its own confidence, and AI rewrites the headline, summary, and category.

Every article ends up in one of three buckets: publish-ready, needs a human, or discard, with the low-confidence middle bucket routed for manual review rather than silently publishing something wrong.

Beyond the pipeline, I designed and shipped full-stack consumer apps for iOS, Android, and web, with customizable push notifications, search and filtering, and analytics. I'm also selling API access to the underlying data to two major data aggregators, so the pipeline itself is a product, not just infrastructure behind the app.

I designed and built this using Figma, Claude Code, the OpenAI API, Google Cloud, Amplitude, and a handful of coordinated agents, making regular calls on model selection, prompt design, evaluation, latency, and cost. Running the pipeline profitably means treating those as product decisions, not just engineering ones.