How VeriLens Works
Learn about the underlying pipeline that evaluates information credibility.
The Credibility Pipeline
To evaluate claims without introducing bias, VeriLens splits the analysis into separate sandboxed components:
1. Extraction & Preprocessing
Input text is parsed. If images are provided, our OCR pipeline extracts text layout elements, identifies central arguments, and filters out boilerplate content.
2. Real-Time Web Context Retrieval
Search queries are generated from the claim key elements. We retrieve facts and articles from trusted news sources, fact-checking databases, and academic indices.
3. Semantic Evaluation & Cross-referencing
The IndoBERT classifier checks linguistic styles typical of hoaxes (e.g., emotional clickbait headers). Concurrently, Gemini cross-references semantic discrepancies between the original claim and verified evidence sources.
4. Verdict Structuring
The final output details the credibility score (0.0 to 1.0) and references the supporting articles. This payload is served back to your system within milliseconds.