Back to case studies
·8 weeks
AI news aggregator with bias scoring · launch to 12k MAU in 8 weeks
Built a news aggregator that pulls multiple sources, fact-checks with AI, and scores bias. Launched to 12k MAU in 8 weeks, zero moderation incidents.
THE PROBLEM
[1/3]
- 01News aggregation is a commodity · needed differentiation via verifiable bias scoring.
- 02LLM-based bias scoring had to be auditable, not opaque.
- 03Multi-source ingestion had to respect every publisher's robots + rate limits.
- 04Launch window was fixed at 8 weeks; no content until the aggregator was live.
THE SOLUTION
[2/3]
- Article-level bias score with explainability · model outputs the 3 sentences that most influenced the score.
- Source verification via redundant retrieval · at least 3 sources must corroborate a claim for a 'verified' badge.
- Rate-limited ingestion pipeline with publisher-specific backoff.
- Evals-as-code pipeline with 400+ labelled articles as gold set.
Technologies
Next.jsPostgreSQLPrismaOpenRouterpgvectorVercel
THE OUTCOME
[3/3]
- 0112,000 MAU in week 8.
- 02Bias-score eval accuracy 91% on the gold set.
- 03Zero moderation incidents in the first 3 months.
- 04Featured in 3 Hungarian tech press outlets.