We built a News Synthesis Engine to fight media bias

Hi HN — we’re Charlie and Sam (two brothers), and we built The Bias, an iOS news app (Android coming soon).

We started The Bias because we found traditional news feeds frustrating: one outlet gives you one framing; opening ten tabs gives you breadth but no synthesis. We wanted a way to understand major events quickly, without spending hours reading or defaulting to a single outlet’s perspective.

For each real-world event we publish a single write-up that makes it clear:

• What coverage converges on (core facts)

• Where perspectives or framing differ

• What’s disputed or still uncertain

• Where specific claims came from (full source list + claim-level traceability for key statements)

• How media around the world reported on the event — which can look very different from domestic coverage. This is especially interesting right now with the situation in Iran, where coverage diverges significantly across regional and international outlets.

How it works (high level):

• Ingest ~5,000–15,000 articles/day from 600+ sources

• Cluster articles by semantic similarity to group reports about the same underlying event

• Extract structured elements (facts, opinions, consequences, speculation, prior context)

• Group and annotate that information

• Generate a first draft with LLMs

• Run the draft through a multi-stage pipeline (coverage completeness checks, contradiction handling, perspective balancing, source traceability, uncertainty labeling, etc.)

• Human review of the final article focused on misrepresentation, hallucination, and missing major viewpoints

We also publish a weekly “rundown” (~25 min read) that summarizes the week, personalized to a user’s interests, using roughly the same process.

The product exists today on iOS and can be used immediately:

• Methodology: https://thebias.co.uk/policies/what-we-do

• About: https://thebias.co.uk/policies/about-the-bias

Free download; limited reading without a subscription. Subscription removes limits and unlocks features like claim traceability and international outlook.

We’re especially interested in feedback on:

• Does this feel faster than reading multiple outlets, or just different?

• What would make you trust (or distrust) a synthesis like this?

• Is claim-level traceability enough, or would you want stronger signals?

• Is this meaningfully different from using Ground News?

• If you think this is a bad idea, we’d genuinely like to know why.

Happy to answer questions about the pipeline, trade-offs, or mistakes we’ve made so far.

— Charlie and Sam