AI story clustering
BGE-M3 multilingual embeddings (1,024-dim, ≥0.78 cosine threshold) detect when 3+ outlets are covering the same story so you see one story, not ten copies.
The story
What it is.
Why we built it.
K cha khabar (के छ खबर?, colloquial Nepali for "What's the news?") is a news intelligence layer for Nepal's fragmented press. It aggregates and clusters Nepali- and English-language headlines from 35+ publishers, then surfaces them as bilingual AI-generated story digests with full publisher attribution and ownership labels.
It's deliberately a headline-and-excerpt aggregator — never republishing full articles. Multilingual clustering uses BGE-M3 embeddings with a 0.78 cosine similarity threshold to detect when multiple outlets are covering the same story. Bilingual summaries are drafted independently per language so the Nepali version reads natively, not as a translation of the English one.
Every publisher carries an ownership label (state, private, independent, non-profit) so readers can weigh perspective. Built for the 1.5M+ strong Nepali diaspora, in-country readers wanting cross-publisher view, and researchers studying the Nepali media ecosystem.