QAnswer now offers more granular source-level traceability for answers pulled from large documents. It highlights precise sentence fragments, anchors them with numbered references, and scrolls directly to their location in the file. Whether working with contracts, specs, policy docs, or multi-section reports, this update reduces the overhead of manual verification and improves transparency across complex datasets.
Improvements
- Granular traceability at sentence-fragment level Numbered reference anchors for source attributionNumbered reference anchors for source attribution - Numbered reference anchors for source attribution Numbered reference anchors for source attribution Numbered reference anchors for source attribution - Direct scroll-to-source navigation within documents Numbered reference anchors for source attribution Numbered reference anchors for source attribution - Inline highlighting of extracted content Numbered reference anchors for source attribution Numbered reference anchors for source attribution
Claude 3.7 Sonnet is now available in QAnswer, fully hosted within the EU. Known for its balanced performance on reasoning, summarization, and structured outputs, it’s a solid option for enterprise use cases. Teams can now select Claude alongside other available models like QAnswer LLM, GPT-4o Mini, or Mistral Medium—choosing the one that best fits their technical and compliance requirements.
Prompt transparency is now available in QAnswer. For every generated response, users can inspect the full prompt structure: system instructions, retrieved documents, and the final output. This feature allows teams to verify how a response was constructed. Legal departments can review the full trace for compliance, AI engineers can debug assistant behavior more precisely, and analysts can check for consistency in responses. The full input-output chain is now visible, making the system auditable and easier to validate in regulated or high-responsibility environments.
QAnswer now supports faceted search. Users can combine natural language queries with structured filters such as file type, author, or date range to refine results. Filters are inferred automatically from the indexed content using both metadata and LLM-based analysis. They adjust dynamically based on the structure and content of the dataset. This enables more targeted queries, improves control over large result sets, and supports use cases that require both flexibility and precision.
QAnswer now supports human takeover, enabling a seamless switch from AI responses to live human agents within the same chat interface. When triggered, the assigned staff receives an email notification with a direct link to join the session and respond manually. This is ideal for workflows requiring human validation, escalation, or oversight—especially in sensitive or regulated contexts. The feature integrates natively and supports hybrid support models out of the box.
Smarter Report Generation: Templates, Structure, and Reusability
v2.0.0
Report generation has been a core use case for QAnswer, particularly in workflows where summaries or structured documents follow predefined formats. This release introduces three improvements to streamline that process. Users can now create reusable templates to standardize output across recurring tasks. External formats—such as Word, Markdown, or HTML—can be imported and applied directly. The system also enforces better control over structure and formatting, ensuring that generated content aligns more closely with the expected layout and reduces the need for manual adjustments.
Smarter Acronym Handling: Retrieval and Prompt Awareness
v2.0.0
Acronym definitions (e.g., CPNP: Cosmetic Products Notification Portal) were previously only used during retrieval, helping the system find relevant sources but not improving how the LLM interpreted the query itself. With this release, acronym data is now also injected into the LLM prompt. This allows the model to understand and respond more accurately to queries that reference acronyms—even when the full term isn’t present in the user’s input. No changes are required in how acronym data is added; the enhancement improves coverage and consistency across both search and generation.
Sri Kalidindi
AI Engineer
Try QAnswer Today For Free !
Upload your data, ask your questions, and get precise answers instantly
.jpeg)
.jpeg)
