Embedding Studio Files

Release Notes

Key New Features:

• pgvector Integration as a Vector Database:

  • You can now leverage PostgreSQL with the pgvector extension for efficient storage, indexing, and searching of vector embeddings. This enables new possibilities for semantic search and other embedding-based tasks directly within your relational database.

• Triton Inference Server Integration:

  • Added support for NVIDIA Triton Inference Server. This allows you to deploy and serve your machine learning models with high performance, scalability, and using industry-standard tools.

• Enhanced Search Capabilities:

  • Implemented both similarity search and exact search. You can now find not only exact matches but also semantically similar items based on their vector representations.

Improvements and New Components:

• Model Management Worker:

  • Introduced a new background worker dedicated to handling model lifecycle tasks such as loading, unloading, versioning, and monitoring. This improves reliability and simplifies model management.

• Data Management Worker:

  • Added a worker to automate data processing and preparation tasks, such as indexing data into the vector database or preprocessing data before model inference.

• Redis-Based Suggestions:

  • Implemented a suggestion generation system (e.g., for search autocomplete or recommendations) using Redis. This ensures high responsiveness and relevance of suggestions.

• Query Parsing:

  • Improved the mechanism for parsing and interpreting user queries, allowing for more flexible handling of complex search patterns and better understanding of user intent.

Experimental Features:

• Continuous Improvement:
  • Added an experimental capability for continuous system improvement (e.g., models or search indices) based on incoming data and user feedback. Please note that this feature is under testing and subject to change.