Here’s what’s new and improved across the platform.

Health Check for Document Utilities
A new health check for the document processing service improves monitoring and system reliability.

Image-to-SMILES Processing on Upload
When supported files are uploaded in enabled environments, Cortex automatically calls the image-to-SMILES service to extract molecular structures from images.

External Embedder Model Support
External embedder models can now be used for embedding and retrieval workflows.

Improved Job Failure Handling
If an embedding job fails, remaining batches now automatically stop processing to avoid unnecessary compute usage.

Controller Timeout Alignment
OpenAI controller timeout settings were updated to align with Cortex system defaults for more consistent request handling.

Refactored Map-Reduce Generation Pipeline
The map-reduce pipeline used in generation tasks has been significantly refactored to improve performance, scalability, and maintainability when handling large inputs.

Automatic Source Embedding on Upload
Documents are now automatically embedded at upload so they’re immediately available for RAG workflows.

Improved Error Messaging
More informative error messages make troubleshooting easier.

Improved Model Loading

Plug-and-Play Rerankers
Reranker models can now be deployed and used within RAG workflows.

Data Fabric Source ID Handling
Corrected per-source Data Fabric ID handling to ensure accurate source attribution.

Map-Reduce Endpoint Updates

Model Server Logging
Model server logs now correctly ship to the OpenSearch dashboard.

GPT and Retrieval Performance improved

GPT response times have been fixed to be in line with previous versions of Certara.AI

Custom Document Templates
Users can now create and use custom templates, allowing greater flexibility when generating documents.

Shift Table Processing
Added support for Shift tables, enabling LLM requests based on their contents.

Prompt Formatting Improvements
Inserted prompts now correctly merge with existing document formatting.

Phoenix Configuration Updates
Phoenix template configurations were updated to align with recent platform changes.

Improved Background Processing
Async thread pools were isolated to prevent blocking between background jobs.

Generated File Naming
Generated narrative files are now named based on their associated subsection for easier organization.

Prompt Request Defaults Updated
Updated frequency penalty defaults for prompt requests.

Prompt Node Sources Field
Prompt nodes can now include a Sources field, allowing workflows to reference specific sources more easily.

Multiple Source Support in Generation Nodes
Generation nodes can now accept multiple sources, enabling more flexible prompt workflows.

Layar Search Node
A new node makes it easier to search for sources directly from Layar.

Persistent Data Fabric Selection
Your selected Data Fabric is now saved in local storage, so it remains selected when refreshing or reopening the application.

Advanced Filtering
Resolved an issue where advanced filtering returned incorrect source counts.