A few months ago, my wife showed me an email newsletter that she had been subscribed to for years. She was annoyed that the newsletter had become filled with thinly veiled ads and basically stopped reporting real news, instead opting for high engagement celebrity rumors. The original purpose of the newsletter was to get informed about the world's happenings every morning without having to go through tons of news sites, be bombarded by ads, or--worse--browse social media.

So she asked me if I could build something for her, and I sort of...took it too far? I built a full news aggregation, summation, and delivery service with fact-checking, bias labeling, and comparisons between how different news organizations report each story. It's called BeamWire (https://beamwire.ai).

So the idea was to go from something like this:

> "Trump alleges 'stunning' Chinese election interference, deep state coverup" -- covered by 9 outlets

to this:

> Verdict: FALSE. The declassified 2021 intelligence community assessment (ICA 2020-00078D) concluded with high confidence that China "did not deploy interference efforts" in the 2020 election. The documents released to support the claim said voting systems "would be difficult to manipulate" (NYT), and even sympathetic coverage found "many documents, no smoking gun" (The Hill).

That's a real fact-check from a recent brief. Each story gets a summary, bias and sensationalism scores, a claim-by-claim fact check, and a comparison of how different outlets framed it: https://cdn.beamwire.ai/briefs/worldInBrief/806/brief.html

I wanted to build something positive with AI. Not something that replaces work humans already do well, but something that does what none of us have time for: reading the same story across a dozen outlets, checking the claims against primary sources, and being honest about what's verified, what's unverified, and what's just framing. It's not perfect. Grounding the fact-checker in live web evidence instead of model memory, so it doesn't "refute" news that's simply newer than its training data, has been one of the hardest problems. But it's already caught things I would have swallowed whole.

There are also some fun additions I added to the service:

1. Tons of other topics such as good news ("The Good Stuff"), sports, medical news, and even a misinformation-debunking show

2. AI podcasts of the briefs with tone choice

3. A podcast visualization using avatars (not AI, if you can believe it!) with sentiment analysis and emotional expression based on what's being said

4. The ability to generate custom one-time or recurring "Beams" on any topic and share them with other users if they like

Most of the content is free but I put some of the GPU-heavy workloads behind subscriptions to cover costs (still working on fair pricing for this).

Happy to answer any questions about how it all works and get feedback on what people would like to see!

Edit: I know this is a bit gray area since it IS a newsletter generator, but you can play around with the "custom beam" tool for free to experience the tech.