SAML Single Sign-On (SSO) for Seamless and Secure Sign-In
Users can now log in to Enterpret via SAML SSO, enabling secure and hassle-free authentication. SAML (Security Assertion Markup Language) is a trusted enterprise-grade protocol, widely adopted for centralized and secure access management.
Performance Boosts: Hyde Latency & Wisdom Queries
Feedback Collection Optimizations
We’ve enhanced the performance of feedback collection for smaller data sets, delivering faster and more accurate results.
Key Improvements:
- Streamlined Retrieval: Simplified feedback data retrieval by eliminating redundant steps.
- Better Query Accuracy: Ensures all records in the set are effectively utilized, optimizing results.
Results:
- Overall Time Reduction:
- Old Average: ~50.03 seconds
- New Average: ~28.85 seconds (42.3% improvement)
- Feedback Collector Time Reduction:
- Old Average: ~20.86 seconds
- New Average: ~3.92 seconds (81.2% improvement)
Hyde Latency Optimization
We’ve made Hyde faster, improving summary collection to enhance the user experience and reduce wait times. By implementing streaming summaries and parallel record retrieval, we’ve achieved significant latency reductions for the Similar Summary Collector:
- P50 Latency: Reduced by 4.5% (11.10s → 10.55s)
- P90 Latency: Reduced by 24.5% (18.79s → 14.26s)
- P99 Latency: Reduced by 28.3% (19.89s → 14.26s)
Fine-Tuned Filter Extractor
Our filter extractor model has been fine-tuned using refined production data. Here’s what makes it stand out:
1. Blazing Speed
- Latency reduced from 6.3 seconds (p50) to an impressive 1.2 seconds (p50) by removing unnecessary reasoning generation.
2. Improved Filter Quality
- Sharper Accuracy: Cleaner, more relevant filters now perform at or above current production standards.
- Smarter Metadata Matching: Reduces noise by avoiding naive text-matching for metadata filters.
- Higher Precision in Reason Filtering: Minimizes irrelevant filter selections, ensuring precise insights.
- Enhanced Recall: Captures more relevant reasons for filter creation.
3. Consistent Error Messaging
- Delivers uniform error messages when context is insufficient, keeping workflows smooth and predictable.