Pinecone provides Assistant for generative AI development

Providing an Assistant

Enterprise investment in developing generative AI tools has exploded over the past two years because of the technology’s potential to make workers smarter and more efficient.

The development process, however, is complex, requiring models and applications to not only be trained with high-quality data but also large volumes of data to reduce the likelihood of inaccurate and even bizarre outputs called hallucinations. The need for both accuracy and volume has made unstructured data such as text and images — which is estimated to make up over 80% of all data — indispensable since it better enables AI tools to comprehend an organization’s operations and adds data volume.

Given that vectors, which are numerical representations of data, can be used to give structure to unstructured data so it can be searched and discovered, vector databases have emerged as a key part of AI development pipelines.

Pinecone Assistant aims to simplify developing AI applications, according to Nathan Cordeiro, the vendor’s principal product manager for generative AI.

By providing APIs and instructions for customizing applications — among other features — Assistant can speed the development process for experienced developers as well as enable less experienced users to create retrieval-augmented generation (RAG) pipelines that discover relevant data and deliver it to generative AI tools.

“For less sophisticated users, we wanted to lessen the learning curve,” Cordeiro said. “Pinecone Assistant was designed to give developers of all skill levels the ability to benefit from a vector database without having to go through all the steps that working with one normally involves.”

Because Assistant not only simplifies AI development for trained developers but also enables non-technical users to build AI applications, it is a meaningful addition for Pinecone customers, according to Stephen Catanzano, an analyst at Informa TechTarget’s Enterprise Strategy Group.

“It’s a significant addition because it streamlines the development of chat and agent-based applications, enabling even non-technical users to build production-grade RAG solutions. [and reduces] time to value.”

In addition to APIs and instructions for customization, specific features of Assistant include the following:

• A chat-based interface that delivers structured responses that include citations so users can check for accuracy.
• Integrations with numerous LLMs to provide users with choices as they integrate proprietary data with models.
• Support for various input/output types such as JSON, PDF, .txt and .docx files.
• Region controls so that users can build in either the U.S. or European Union.
• The ability to handle workloads up to 100 times larger than when Assistant was first unveiled in public preview.

Perhaps most valuable to users are the APIs and instructions for customization, according to Catanzano.

“The APIs ensure accurate, grounded outputs with citations, while custom instructions allow fine-tuning assistants for specific use cases, making them highly flexible and reliable,” he said.

The future

While no longer in preview, Pinecone plans to continue improving Assistant to better enable users to develop generative AI applications, according to Cordeiro.

In particular, the vendor plans to add integrations with new data sources, support more unstructured data types such as images and add new APIs that make it easier for users to customize Assistant to adapt to different domains.

“You can expect to see further investment in the versatility of Assistant,” Cordeiro said.

Investing in versatility would be wise for Pinecone, according to Petrie.

Broad-featured database vendors that handle tables and knowledge graphs are adding vector search and storage capabilities. To remain competitive, Pinecone may have to expand beyond being a vector database specialist.

“Broader databases prepare and deliver diverse data inputs to applications that contain multiple model types — GenAI language models and predictive ML,” Petrie said. “Pinecone should extend its capabilities to better support these diverse data and model types.”

Catanzano likewise suggested adding features that expand Pinecone’s versatility, including support for more LLMs, prebuilt templates for industry-specific applications and more input/output formats.

“These enhancements would broaden its applicability and further ease adoption for diverse use cases,” he said.

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.