Product Overview: EvaDB
EvaDB is a comprehensive, open-source framework designed to integrate AI capabilities directly into database systems, enabling software developers to build AI-powered applications with ease and efficiency.
What EvaDB Does
EvaDB allows developers to add AI-driven features such as regression, classification, image recognition, and question answering to their database applications. It supports the execution of AI queries over data stored in existing SQL and vector database systems, leveraging pre-trained models from prominent AI engines like Hugging Face, OpenAI, PyTorch, and YOLO.
Key Features
1. AI-Enhanced Query Language (EvaQL)
EvaDB uses a high-level, declarative query language called EvaQL, which is derived from SQL. This language enables developers to write AI queries that are similar to traditional SQL queries but incorporate AI models as function calls. This simplifies the development of AI-powered apps by reducing the complexity of integrating AI models into database queries.
2. User-Defined Functions
Developers can register custom user-defined functions that wrap around any AI model or Python library. These functions can be used in the SELECT and WHERE clauses of queries, allowing for the integration of custom deep learning models into the database.
3. AI-Centric Query Optimization
EvaDB includes a novel query optimization framework tailored for AI queries, inspired by the Cascades-style extensible query optimizer used in SQL database systems. This optimization includes techniques such as function caching, sampling, cost-based operator reordering, and parallel query processing, which significantly reduce query execution time and the cost of running AI models.
4. Performance Enhancements
EvaDB accelerates AI queries through several optimizations:
- Function Result Caching: Reuses results of expensive AI function calls.
- Batching: Reduces token usage and costs associated with large language models (LLMs).
- Parallel Processing: Better utilizes CPUs and/or GPUs to save time and money.
- Predicate Re-ordering and Push-down: Accelerates queries over both structured and unstructured data.
5. Support for Multiple AI Engines
EvaDB provides first-class support for various AI engines, including PyTorch, Hugging Face, YOLO, and OpenAI, making it versatile for different AI tasks such as NLP, vision applications, and more.
6. Ease of Use and Integration
EvaDB is fully implemented in Python and can be installed via pip. It offers a simple and intuitive SQL API that allows even beginners to use AI models in their database queries with just a few lines of code.
Functionality
- Connecting to Database Systems: EvaDB can be connected to existing SQL and vector database systems using CREATE DATABASE and CREATE INDEX statements.
- Writing AI Queries: Developers can write SQL queries that include AI functions to obtain inference results from pre-trained models or custom models.
- Custom Model Integration: Developers can bring their custom AI/ML models into EvaDB using the CREATE FUNCTION statement, allowing for a wide range of AI applications to be automated within the database system.
In summary, EvaDB simplifies the development and deployment of AI-powered applications by providing a powerful SQL API, extensive support for various AI engines, and advanced query optimization techniques. This makes it an ideal solution for developers looking to integrate AI capabilities into their database-driven applications.