Release Notes v1.2
Features
1. Deduplicated Citation Sources
This release introduces deduplicated source tracking alongside verbatim citations, improving clarity and auditability of AI-generated responses.
What’s new:
-
Each response can now display both the verbatim citation and its source origins.
-
Reduces repeated references and improves traceability across large document sets.
-
Enhances user confidence in understanding exactly where information came from.
2. LLM Hallucination & Confidence Scoring
We’ve upgraded our hallucination detection framework with a more advanced LLM-powered verification process.
Key enhancements:
-
The system automatically rechecks sources and generated answers using a LLM.
-
Provides a reasoning summary explaining how the answer was validated.
-
Includes a confidence score highlighting likelihood of hallucination.
-
Improves overall reliability and transparency of generated content.
3. Multi-Query Support
This release adds support for processing multiple related queries within a single request, enabling richer and more context-aware responses.
What you’ll see:
-
Better coherence across multi-part questions.
-
Reduced need for repeated follow-up queries.
-
More complete, holistic outputs for complex user tasks.
4. Improved document ingestion and chat responses
We’ve strengthened document ingestion, parsing, and transformation workflows - to improve chat response quality.
Improvements include:
- More accurate extraction of content from PDFs, Microsoft Word and Excel files.
- More reliable preprocessing for downstream LLM operations.