DEV Community

mslugga35
mslugga35

Posted on • Originally published at getdawnbrief.com

How I Used AI to Build a Personalized News Filter That Actually Saves Time

The Problem: Too Much Signal, Not Enough Filter

Every morning I'd open my laptop and immediately feel behind. TechCrunch had 18 new posts. My RSS feed had 200+ unread items. Three newsletters I'd subscribed to months ago were sitting unread. I knew somewhere in all that noise were 3-4 stories that were genuinely relevant to what I was building — but finding them cost me almost an hour.

The irony: I was building software to save people time while spending my mornings doing something a machine should obviously be doing for me.

What I Built

DawnBrief is an AI-curated morning digest. The core mechanic is simple: you select your interest categories (SaaS, AI/ML, infrastructure, fintech, developer tools, product management — there are about 15), and every morning before 7 AM, it delivers a clean email with only the stories that match.

Under the hood:

  • Pulls from 50+ curated sources (tech news, research papers, SaaS-specific blogs, funding databases)
  • Uses an LLM to score each story's relevance against a user's interest profile
  • Deduplicates across sources (you don't need to see the same OpenAI funding round in four different framings)
  • Generates short, opinionated summaries — not just headlines

The Technical Angle

The scoring pipeline runs on a cron job. Each article gets embedded, compared against a per-user interest vector, and scored. Stories above a threshold make the digest. Stories below get dropped.

The interesting engineering challenge was calibration. Early on, the AI was too conservative — users were getting 2-3 stories when they wanted 8-10. Too liberal and it started including tangentially-related noise. I ended up with a hybrid approach: hard category filters first, then LLM relevance scoring on the filtered set, then a "top N" cap with a floor to prevent empty digests.

Delivery is straightforward — templated email, sent via transactional email API, personalized per user.

What I Learned

The biggest surprise was how much users cared about consistency. Getting it in their inbox before they reached for their phone in the morning mattered more than having perfect relevance. Reliability is the feature.

Second surprise: people don't want to configure things. The first version had a detailed interest selector with 40+ options. Almost everyone picked 3-5 broad categories and never touched it again. I simplified the onboarding down to a single screen.

Try It

DawnBrief is live at https://getdawnbrief.com — $19/month, cancel any time. If you're a developer or founder who's also drowning in tech noise, I'd genuinely love to hear what you think.

What sources or categories are missing from your morning routine? Let me know in the comments.

Top comments (0)