Kashef
Client
Liminal
Industry
Fact-Checking & Verification
Year
2026
The Challenge
In today's information landscape, verifying news and claims requires cross-referencing multiple sources, assessing credibility, and distinguishing between recent and outdated information. Manual fact-checking is time-consuming and difficult to scale, while automated systems often lack the nuance needed for reliable verification.
Our Solution
Liminal developed Kashef, a hybrid fact-checking platform that combines an internal knowledge base with live internet search. The system processes claims through an intelligent pipeline: extracting verifiable statements, identifying entities (locations, organizations, people), building targeted search queries, checking against existing evidence in the knowledge base, and when needed, conducting live searches to gather fresh sources.
Results
Kashef empowers newsrooms, research organizations, and information platforms to verify claims at scale while maintaining accuracy and transparency. By combining institutional knowledge with real-time search, the platform delivers reliable fact-checking that adapts to evolving information landscapes.
How It Works
The verification workflow begins when a user submits a news article or claim. The system extracts individual claims and identifies key entities mentioned. It then searches the internal knowledge base first, measuring both similarity and recency of existing evidence. If the evidence is insufficient or outdated, the system transitions to live search, scraping and analyzing relevant sources from across the web. All evidence is ranked by relevance and credibility before being compared against the original claim. The final output provides a verification verdict supported by cited sources and confidence scores.
Key Capabilities
- Automated claim extraction from news articles
- Entity recognition (people, places, organizations)
- Hybrid search: internal knowledge base + live web search
- Evidence ranking by relevance, credibility, and recency
- Source citation and confidence scoring
- Scalable verification pipeline