Last week at the FIBEP Tech Day conference, our Jiri Hana spoke about the practical challenges of building RAG systems for news publishers and media monitoring companies.
RAG for news monitoring is a different beast than the typical tutorial examples. We deal with datasets of hundreds of millions or even billions of articles.
A few things we’ve learned at Geneea:
- Smart Filtering: Using filters over metadata and named entities to narrow down the dataset improves quality tremendously. But users rarely do this manually, so we derive the filters automatically.
- Accuracy over prose: For power users, correctness and clear source attribution are far more important than how nicely the answer is written.
- Attribution: It’s not just about what was said, but who said it — distinguishing between the journalist, the publication, and a quoted politician.
For those interested in the technical details here are the slides, but they are only a summary — the most interesting parts were the live discussions.
Hiking in Tenerife
Also, Tenerife is a great place for hiking: one can climb Teide (the highest mountain in Spain), explore the moon-like landscape around it, the rainforest in the North, or Masca — the “Machu Picchu” of Tenerife in the South.
