Incorporated New Developer Options: Enhanced functionality with the addition of advanced configuration settings and tools.

Implemented Various Bug Fixes and Improvements: Addressed known issues and optimized performance for a smoother user experience.

Significant Improvements in Document Parsing: Enhanced accuracy and efficiency in extracting and processing data from various document formats.

Enhanced Large Source File Import Process: Improved handling and efficiency for importing extensive source files, reducing load times and increasing reliability.

Resolved Issues with Narrative Preview Generation: Improved stability and accuracy in creating narrative previews, ensuring more reliable and precise outputs.

Migrated to PKCE-Based Login: Improved security and compliance by implementing Proof Key for Code Exchange (PKCE) for OAuth 2.0 authorization flows.

Added Default Sort Values to All Tables: Ensured consistent and efficient data organization by implementing default sorting parameters across all database tables.

Save and Display Most Recent Modification Date for Documents: Updated document metadata to prioritize the latest modification date over the creation date, ensuring more accurate tracking of document changes.

Implemented Various Bug Fixes and Enhancements: Addressed multiple issues and optimized system performance for a more stable and efficient user experience.

New Create Embeddings UI: Adds a UI for creating embeddings with settings exposure, progress tracking, failure alerts, and retry options.

Added URL Scraping Capability for GPT Generation Requests: Enabled the extraction of content from URLs to enhance the accuracy and relevance of GPT-generated responses.

Introduced New GPT Retrieval Endpoints: Added /GPT/retrieval/search and /GPT/retrieval/generateVectors endpoints to improve search capabilities and vector generation for more efficient data retrieval and processing.

Resolved Text Classification Issues for Large Jobs: Improved the accuracy and efficiency of text classification processes for handling extensive datasets.

Corrected Datatype for Rerank Score: Ensured accurate data handling by fixing the datatype for rerank scores, improving the reliability of ranking algorithms.

Resolved Non-JSON Response from /GPT/generate/async Endpoint: Ensured consistent JSON formatting for asynchronous GPT generation requests, improving data interoperability and error handling.

Enhanced Embedding Creation Process: Improved the accuracy and efficiency of embedding generation, resulting in better data representation and retrieval performance.

Improved Error Handling and Responses for GPT Requests: Enhanced the robustness and clarity of error messages and responses, ensuring more reliable and user-friendly interactions with GPT services.

API Documentation Fixes: Corrected inaccuracies and updated content to ensure clear, accurate, and comprehensive API documentation for developers.

Implemented Comprehensive Bug Fixes and Enhancements: Addressed multiple issues and optimized system performance to ensure a more stable, efficient, and user-friendly experience.

User Modification Restrictions with SSO Enabled: When Single Sign-On (SSO) is active, users are restricted from changing their name, username, or email address to ensure consistency and security.

New Configurable Token Expiration: Set custom token expiration times for enhanced security and flexibility.

Admin User and Seat Assignment Pre-SSO: Enables admins to create users and assign seats prior to initial SSO login, streamlining user onboarding and ensuring immediate access to necessary resources.

New Admin Communication Endpoints: Introduced new API endpoints enabling admins to send emails to all users, facilitating efficient and centralized communication.

Classification Model Count Fix: Resolved an issue causing incorrect counts post-training, ensuring accurate model performance metrics and reliable classification results.

Dynamic Token Limit Adjustment: Automatically modifies the maximum token limit according to the available context length and input prompt size, optimizing resource utilization and ensuring efficient processing.

New Model Support: Introduced compatibility for Llama 3.2 (Llama Guard) and Llama 3.3 models, enhancing model versatility and performance capabilities.

Custom Embedding Model Specification: Enables users to select different embedding models hosted on the instance for the /createEmbeddings, /generate, and /retrieval endpoints, providing flexibility and optimizing model performance for various tasks.

Custom LLM Seed Setting: Allows users to specify a custom seed for their language models, ensuring reproducibility and consistent results across different runs.

New Retrieval Endpoints: Introduce RAG-based search capabilities on pre-indexed documents and enable vector generation for custom text, enhancing search precision and flexibility.

Embedding Models Display Endpoint: Introduces a new endpoint to list all embedding models currently hosted on an instance, providing visibility and management capabilities for available models.

LLM-Based Tokenizer Integration: Replaced the NLP tokenizer in the Sentence Splitter with an LLM-based tokenizer to ensure uniformity in token count computation, enhancing accuracy and consistency.

Enhanced Error Handling in Document-Sectioning: Implemented additional error handling for edge cases where no chunks are available for expansion, ensuring robust and reliable document processing.

Weaviate Index Validation: Added validation and error checks to ensure documents exist in the Weaviate index when using the /retrieval/search endpoint with document IDs in search filters, enhancing search reliability and accuracy.

Enhanced Logging Capabilities: Super-users and admins can now enable Detailed Logging for both certara-gpt and Cortex pods, improving monitoring and troubleshooting across these components.

GPT Terminology Update: We have updated our GPT generation terminology, replacing 'system' with 'assistant' to better reflect the AI's role. This change enhances clarity and consistency in our documentation and interfaces. Note that 'system' will still be used in model prompts, but the response message type will now be called 'assistant'.