Application developers are in a race to build generative AI applications that can understand and augment their business-specific context to solve their needs. Vector search is the de facto solution for searching and retrieving this semantic context. Many vendors offer vector search capabilities, but developers consistently run into challenges with hallucinations and performance issues, lack of real-time context, and complexities in development and integration with GenAI ecosystems. These hurdles can significantly lengthen the time to production.
That’s why we’re excited to announce a public preview of the new DataStax Astra DB vector experience, one that shortens and simplifies the GenAI application development journey. Here’s what’s new:
- A new user experience - A new UX that enables you to easily load and explore your vector and non-vector data for quick analysis and relevance improvements.
- Native clients - Intuitive Python, JavaScript/TypeScript, and Java clients to accelerate your application development.
- GenAI integrations - Seamless integration with GenAI ecosystems including LangChain, LlamaIndex, and Google Cloud Vertex AI, to enable your applications to leverage the benefits of the AI ecosystem.
- A new API - An optimized Astra DB API that’s been updated with schemaless document-style interactions to reduce the burden of bespoke database data modeling for GenAI application development.
- JVector -A best-in-class algorithm to turbocharge the performance of production GenAI applications.
A new user experience
The new UX enables you to upload both vector and non-vector datasets. Astra DB with vector offers unique capabilities to explore your data without any code, and easily view usage and performance metrics. You can also perform cosine similarity analysis, and add and remove metadata filters for quick analysis and relevance tuning.
Native clients
Enjoy the power of Apache Cassandra for your GenAI applications without having to learn Cassandra or CQL (Cassandra Query Language). Our new native clients for Python, JavaScript/TypeScript, and Java enable you to access features in the programming language you’re most familiar with.
Here’s an example of how easy it is to initialize and create collections using the Astra DB Python native client. Just install the library:
pip install astrapy
Then Import and use:
import os
from astrapy.db import AstraDB
ASTRA_DB_APPLICATION_TOKEN = os.environ.get("ASTRA_DB_APPLICATION_TOKEN")
ASTRA_DB_API_ENDPOINT= os.environ.get("ASTRA_DB_API_ENDPOINT")
# Initialization
db = AstraDB(
token=ASTRA_DB_APPLICATION_TOKEN,
api_endpoint=ASTRA_DB_API_ENDPOINT)
# Create collection
test_collection = db.create_collection(collection_name="vector_test", dimension=768)GenAI integrations
Large language model (LLM) orchestrators like LangChain and LlamaIndex provide the interface between LLMs and your GenAI application.They provide composability for AI agents to integrate with the environment and ecosystem. You can easily use the new Astra DB experience with LangChain and LlamaIndex. Moreover, you can leverage the strength of hyperscaler services such as Google Cloud Vertex AI with Astra DB.
Here is an example of using the new Astra DB vector experience with LangChain:
from langchain.vectorstores import AstraDB
Running a similarity search is as simple as:
results = vstore.similarity_search("Our life is what we make of it", k=3)Astra DB API
With the new schemaless Astra DB API, you can make simple and standard JSON document interactions with both your vector and non-vector data. The high-performance API provides an abstraction that is agnostic to bespoke database-specific data modeling. This API powers the new clients, giving you unified capabilities irrespective of the programming language you use.
JVector
JVector is a state-of-the art vector search engine that provides benefits such as speed, memory-efficiency, disk-awareness, concurrency, and ease of embedding. These are attributes that power real-time data for real-time use cases in Astra DB. JVector uses optimized graph algorithms inspired by DiskANN, meaning that vectors can be compressed using product quantization so that searches can be performed using the compressed representation that is kept in memory. This surpasses the scale and speed at which common vector engines like HNSW (Hierarchical Navigable Small World) operate. Learn more in the JVector GitHub repo.
Astra DB vector in action
DataStax customers across industries rely on Astra DB to build generative AI applications. Here’s what two of them had to say about using Astra DB vector.
“We've relied on Astra DB for its efficient data handling capabilities, low latency, and quick read times,” said Matthew Hirst, head of engineering at Hornet, a leading social networking application for the queer community. “Our choice of DataStax's vector search seamlessly integrates with our current cloud infrastructure. By harnessing GenAI for personalized matching, we're advancing meaningful and safe connections within the LGBTQ+ community while reducing databases and dependencies for us to manage.”
"Restworld revolutionizes hospitality recruitment with DataStax Astra DB and vector search, enhancing candidate matching,” said Edoardo Conte, Restworld founder and CTO. ”With 90,000+ registered workers in Italy on our platform, we anticipate a significant efficiency boost and a potential 5X improvement in candidate matching, using collaborative filtering algorithms and vector capabilities inspired by tech giants like Netflix and Amazon. Our meticulous job profiling and semantic search are driving this innovative recruitment approach."
Get started with Astra DB today
Leverage the power of simplicity, scale, and the GenAI ecosystem with Astra DB; build your GenAI application with the new vector experience. Sign up for DataStax Astra DB, and select “Enable Preview” to get started.
