Starburst intros AI assistant to boost analysis, exploration

Assisting exploration and analysis

For decades, data exploration and analysis were limited to trained data scientists and analysts. Complex coding, data literacy and proficiency in statistics were among the requirements. Natural language processing promised a means to more widespread use of analytics, but technological limitations prevented NLP from making analytics more accessible to non-technical users.

Although I would not call the ReAct framework 'advanced reasoning' by today’s standards, it is a meaningful improvement over most natural language-to-SQL products, which are essentially one-shot translators. If AIDA genuinely cycles through refinement, that's worth taking note of.

OpenAI's November 2022 launch of ChatGPT changed that, enabling freeform NLP for the first time.

Now, like some of its peers, Starburst is improving on its initial generative AI query tools with more contextually aware AI-powered data exploration and analysis capabilities. Customer feedback drove Starburst to develop AIDA, according to Justin Borgman, the vendor's co-founder and CEO.

"Customer demand was the primary driver, specifically the frustration of waiting weeks for dashboards that business users didn't fully trust," he said. "But the deeper insight was that AI only becomes useful in the enterprise when it operates on real, governed business context."

Borgman added that while some vendors already provide AI assistants, Starburst is building and refining AIDA because customer demand for more advanced capabilities is increasing.

"They need AI that works across their data estate, understands business context, and integrates into their broader tool ecosystem. … That's the problem our architecture was built to solve," he said.

In addition to the ReAct framework, persona-based outputs and support for multiple LLMs, AIDA can be rebranded by customers with their own logos and color schemes so users' experiences with the AI assistant look and feel like their other workflows.

Beyond current features, Starburst plans to add the following tools in the coming months:

• AIDA Studio, a layer that enables developers to connect the AI assistant to external systems, incorporate unstructured data such as text and images and build multi-step workflows.
• Orchestration capabilities that power agentic AI workflows, enabling AIDA to reason through multi-step processes to autonomously act across connected systems.
• A governance layer beyond existing data access controls that provides guardrails for what users can ask and the AI tool can return, governing the interaction itself rather than just the underlying data to prevent exploration of restricted or sensitive subject matter.

Kevin Petrie, an analyst at BARC U.S., noted that while many vendors provide AI assistants, organizations struggle to benefit from such capabilities because of poor governance and disconnected data. Given that Starburst's platform addresses those obstacles, he noted that AIDA is a valuable addition for Starburst's users.

"Starburst helps close the adoption gap by providing governed access to distributed datasets," Petrie said. "So rather than making costly migrations or compromising sovereignty, Starburst customers can apply powerful agentic reasoning to data wherever it sits. The agent simply uses federated querying across platforms."

In addition, persona-specific capabilities could help distinguish AIDA from other AI assistants, Petrie continued.

"AIDA differentiates itself with persona-specific agents that guide natural language conversations based on the technical or business acumen of the human user," he said. "Starburst also improves governance and scalability by standardizing its queries on consistent, reusable and modular data products."

Farmer, however, cautioned that AIDA's current lack of guardrails to govern interactions -- that their addition is in the coming months rather than the initial preview release of the AI assistant -- is a concern.

"The fact that guardrails are 'coming soon' unintentionally tells you everything you need to know about their priorities," he said. "Reasoning first, governance later -- that's not good."

Looking ahead

Beyond the next wave of features for AIDA, Starburst's product development plans include improving its enterprise context layer, according to Borgman.

"This is what separates AI that enterprises can actually deploy from AI that hallucinates in an incomplete context," he said.

In addition, Starburst plans to upgrade its governance capabilities by improving access controls and the protections that safeguard data sources and agentic interactions, Borgman added.

"We're building Starburst to be the operating layer where enterprise AI actually works, not by asking enterprises to move their data into our platform, but by bringing intelligence, governance, and context to data wherever it already lives," he said.

Petrie noted that Starburst's focus on federated data environments rather than single-source data lakehouses is in line with rising trends -- especially in Europe -- related to data sovereignty and managing data in place. Going forward, he suggested that Starburst focus on the extensiveness of AIDA's unstructured data exploration capabilities.

"I'll be interested to see how Starburst broadens AIDA support for unstructured data, including the documents, emails, and images that contain critical context for GenAI models and agents," Petrie said. "I think it can do more to help data and business teams curate unstructured content with tagging, classification, and integration into context engineering processes."

Farmer, meanwhile, noted that Starburst's focus on distributed data environments is valuable, with cross-domain analysis potentially even helping differentiate AIDA from other AI assistants. But before Starburst adds any other features, he noted that it needs to get AIDA's guardrails in place.

"They need to get that fixed before another round of feature development," Farmer said.

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