Using a spirit level while building a garden wall from English Cotswold stone and cement mortar. … [+]
Software has a lifecycle. Our apps are code-based entities that are created during what is known in the IT industry as the Software Application Development Lifecycle (SDLC). Code is written, tested and debugged, released to a live production environment, and then extended, enhanced, maintained, and expanded. Over the long term, some applications will transition to legacy status (a term that is usually used pejoratively, but at least these are still-functioning apps), and some will be refactored, retired, decommissioned, or retired.
We could add more steps and phases here, but the point is clear: software has a lifecycle. As the tech industry enters its current artificial intelligence renaissance, we can even talk about an AI lifecycle.
What is the AI lifecycle?
The AI lifecycle starts with infrastructure, which includes everything from the large language models used to the logic of the AI engines and even the analytics capabilities of the cloud computing services that development teams sign up for. It also includes the database provisioning process, the networking and connectivity layers that join microservices and interface with external applications and services through application programming interfaces. At a high level, the AI lifecycle focuses on compute processing, the throughput of the data itself, and the delivery of intelligence to end users through chatbots and other smart software features. Again, this is not an exhaustive list, but it gives some idea of the circular flow that exists.
DataStax looks to make much of the back-end part of the lifecycle equation easier, allowing developers to focus on application development rather than infrastructure management. Now positioning itself as an AI platform company, DataStax is focused on generative AI development with extensions that make search extension generation easier. Highly popular with AI purists who want to make AI real, RAGs provide a means to inject externally approved data into the AI process. Often that data belongs to a dataset owned by the company using the deployment.
The Truth About Langflow
In April of this year, DataStax acquired Langflow, an open-source visual framework for building RAG applications. Now, DataStax has released Langflow 1.0, which includes a version of Langflow hosted on the DataStax Cloud platform. The technology has a drag-and-drop interface and numerous integrations with the most popular generative AI tools, including LangChain, LangSmith, OpenAI, Hugging Face, and Mistral, allowing developers to configure, swap, and compare all the major large-scale language model and embedding providers.
According to the company, this gives developers the flexibility to compare results with different providers. Developers can make big changes faster without having to learn new APIs or recode their applications. As part of the Langflow 1.0 open source release, developers can now use LangSmith's observability service to track application responses to create more relevant and accurate LLM-based applications.
“The generative AI stack and its lifecycle is a large, complex chunk of technology that many are trying to understand. We're focused on staying true to our roots so developers can focus on what they do best: building and developing, rather than worrying about application infrastructure,” said Ed Anoff, chief product officer at DataStax. “From the release of Langflow in Astra, [DataStax’s serverless, multi-model database]”By bringing together the largest ecosystem of embedded providers in one place, we are delivering on our promise to make generative AI application development as fast and easy as possible, enabling organizations to rapidly deploy apps into production and see immediate benefits.”
Ready-to-use AI data
As part of its current news stream, DataStax also mentioned a new partnership with Unstructured, a no-code platform and cloud service that turns nearly any document, file type, or layout into LLM-ready data and “stands up” generative AI data pipelines, from transforming and cleaning to generating vector database embeddings. The partnership is designed to make it easier for enterprises and developers to prepare their enterprise data for AI by handling data ingestion and chunking (the use of so-called data chunking algorithms allows data to be split into smaller, more manageable chunks) across data types such as PDF, Salesforce, Google Drive, etc. for use in AI applications.
Again, the company aims to take on more of the back-end lifecycle heavy lifting.
At this point, AI developers can ingest large datasets and common document types by quickly converting them into vector data. This new integration enables them to quickly write these embeddings into Astra DB to perform associated generative AI similarity searches.
Infrastructure Offload
“Developers want to take advantage of generative AI and discover how they can use it in their applications. They're under pressure from the business to figure out what they can accomplish and what options they have to innovate,” says DataStax's Anuff. “But because this space is so new, there's constant development and new releases. New versions of tools are released, and these updates can affect or break applications, making it hard for developers to keep up. Engineers are essentially spending more time building infrastructure than building software and solving problems.”
To overcome these hurdles, we are finding that software teams are looking at ways to remove or reduce the infrastructure overhead around generative AI, so they can focus on business problems rather than IT issues and ensuring alignment.
“For the majority of developers working with generative AI, reducing or eliminating the integration overhead around their applications allows them to focus on business logic and application design, creating the most value for the business. For developers implementing new techniques like RAG, getting accurate performance data across the entire application lifecycle can help find bottlenecks that are impacting response times and show where different options can improve results,” explains Anuff.
Shaping the AI life cycle
Every company building generative AI today is looking for the most effective way to implement their toolsets and extensions (and usually RAGs) into their applications. At this point, when the technology is (probably) still in its infancy, being able to split the AI development process into front-end, back-end, upper-tier, and lower-tier elements across the AI lifecycle helps you think about the critical building blocks of AI in more manageable terms.
Of course, you can think of the AI lifecycle as cyclical, but when building AI, technologists typically refer to it as an end-to-end process, which in this case means from infrastructure to users. If we can understand this concept, we can stand on a firmer foundation of intelligence despite the AI hype that surrounds us.
