Charlie

Charlie implements a collaborative filtering and content-based recommendation engine that learns user reading patterns over time to surface relevant stories. The system tracks article engagement signals (clicks, dwell time, completion rates) and maps them against user-declared interests and implicit preference signals to rank and filter incoming news stories from partner sources. This creates a dynamically-weighted feed that adapts as reading behavior evolves, rather than applying static keyword matching or manual curation rules.

Charlie maintains a vetted network of news sources (publications, wire services, independent outlets) from which it aggregates stories. The integration layer normalizes article metadata (title, byline, publication date, category tags) across heterogeneous source APIs and feeds (RSS, JSON APIs, web scraping) into a unified internal schema. Source quality and coverage diversity are managed through editorial curation rather than algorithmic inclusion, ensuring baseline journalistic standards while limiting the breadth of available sources.

Charlie provides a minimal, ad-free reading interface that prioritizes article content over navigation chrome, ads, or recommended-content sidebars. The interface silently tracks engagement metrics (scroll depth, time-on-page, reading speed, completion status) via client-side JavaScript instrumentation without explicit user action, feeding these signals back to the personalization engine. The design philosophy prioritizes reading experience over monetization, with no interstitial ads, paywalls, or tracking pixels from third parties.

Charlie allows users to declare and manage interest categories (e.g., 'Technology', 'Climate', 'Local Politics') which serve as explicit preference signals for the personalization engine. The system maps incoming articles to these user-defined categories using NLP-based topic classification (likely keyword matching, TF-IDF, or lightweight ML models) and uses category-level preferences to weight feed ranking. Users can adjust interest weights (e.g., 'Technology: high priority', 'Sports: low priority') to directly influence feed composition without relying solely on implicit reading signals.

Charlie continuously polls partner news sources (via RSS, APIs, or scheduled scraping) to ingest new articles, typically with a refresh cadence of 15-60 minutes depending on source priority. The system implements duplicate detection (likely using content hashing, title similarity, or URL canonicalization) to identify when multiple sources cover the same story, clustering them together and attributing coverage to all sources. Feed freshness is maintained by prioritizing recent articles in ranking, ensuring users see breaking news and developing stories without stale content dominating the feed.

Charlie maintains a persistent user profile that stores interest declarations, engagement history, and personalization weights across sessions. The profile is stored server-side (likely in a relational database) and synchronized with client-side session state, allowing users to maintain consistent personalization across devices and sessions. Profile data includes interest categories, reading history (article IDs, timestamps, engagement metrics), and derived preference weights that feed the ranking algorithm. Users can view and manually adjust their profile (interests, weights) to correct or refine personalization.