The signal-vs-noise problem is universal in AI products. You nailed the core insight — most data isn't actionable, and AI's job should be filtering, not just presenting.
We deal with a similar challenge at AnveVoice (voice AI for websites). When a user speaks to our voice agent on a website, the AI has to filter intent from conversational noise in real-time — deciding in under 700ms whether someone wants to book an appointment, ask a question, or just browse. Getting that classification wrong means taking the wrong action on the page.
Your approach of using AI to rank and surface only what matters is exactly right. How are you handling false negatives — cases where the AI filters out a headline that actually ends up moving the market?
Cool build. The "no distractions, just signals" philosophy resonates with how we think about voice UX too.