Delphina Raises $15M to Automate Enterprise Data Context with AI Agents

SAN FRANCISCO – November 12, 2026 — Every executive knows the frustration of the data ticket queue. A simple question about quarterly churn or regional sales performance gets filed, assigned, and processed, often taking weeks to return an answer that is already stale. Delphina, a San Francisco-based startup, is betting that artificial intelligence can completely eliminate this bottleneck. Today, the company announced $7.5 million in new funding—bringing its total capital to $15 million—to scale its newly launched Context Factory. Co-led by Radical Ventures and Costanoa Ventures, the round includes high-profile participation from Stanford AI luminary Fei-Fei Li, Nobel laureate Guido Imbens, and early engineers from Anthropic and OpenAI.

Bypassing the Data Ticket Queue

The premise of the Context Factory is deceptively simple: AI analytics are only as reliable as the business context underpinning them. In most enterprises, this context—which tables are trustworthy, how specific marketing metrics are defined, and the undocumented institutional knowledge held by veteran analysts—is built by hand and decays rapidly. The new platform hands this maintenance over to autonomous AI agents. These agents ingest existing data models, dashboards, and query logs to build a working baseline in hours. As employees ask questions, the system identifies gaps, drafts fixes, and routes significant changes to human domain experts for approval.

The impact on organizational behavior is already visible. According to the firm, 73% of questions on its platform now originate from business teams rather than centralized data engineers. More strikingly, chief executives at more than half of their customer organizations are active, direct users.

At Substack, the shift has been systemic. Over 70% of the newsletter platform's employees actively use the tool to query company data directly. "We want to give our people superpowers with AI. Delphina does that for data," said Chris Best, CEO of Substack. "Now our teams get accurate answers in seconds instead of waiting weeks in a queue. It's changed how we make decisions."

This acceleration is echoed across a customer roster that includes LATAM Airlines, Gen Digital, Upstart, and Coursera. Industry insiders note that one healthcare staffing platform, Medely, used the system to draft 75% of its third-quarter board deck the day after the quarter ended—a process that historically consumed over a month. For Gen Digital, the platform facilitated over 9,000 AI analytics conversations per month within its first six months of deployment.

From Uber Michelangelo to the Last Mile of Enterprise Data

The architecture behind this autonomous semantic layer is rooted in the founders' experiences scaling some of the most complex data infrastructure in the tech industry. CEO Jeremy Hermann previously built and led Uber’s Michelangelo machine learning platform before co-founding Tecton, a feature store company later acquired by Databricks. President Duncan Gilchrist led global pricing data science at the ride-hailing giant.

"At Uber, everyone, including the CEO, was hands-on with data. That direct line to the numbers was a real advantage, but it took an army of data engineers to keep it from collapsing into chaos," said Hermann. "The Context Factory gives every company that same advantage. Agents do the heavy lifting, and data teams become the knowledge captains who decide how the business sees its numbers."

The transition from a manual semantic layer to an agentic one represents a significant evolution in the startup's product strategy. When initially funded in late 2023, the firm was positioned primarily as an LLM-powered copilot for data scientists. The explicit launch of the Context Factory marks a pivot toward the foundational layer of enterprise intelligence. By focusing on the automated upkeep of institutional knowledge—the "last mile" of data reliability—the founders have attracted investors who specialize in applied AI infrastructure.

"Most companies are data-obsessed but struggle to become data-driven," noted Gilchrist. "When everyone can get trusted answers, everyone suddenly has questions. Ops, sales, marketing and finance teams are now the heaviest users of Delphina, and they keep coming back."

The backing of figures like Fei-Fei Li and Guido Imbens is particularly telling. Imbens's Nobel-winning work in causal inference underscores a broader credibility revolution in empirical economics—a demand for rigorous, trustworthy data insights that is now bleeding into corporate governance. Li’s involvement signals deep technical validation of the agentic approach to parsing complex, unstructured institutional knowledge.

The Battle for Enterprise Context Against Incumbents

Despite the impressive early traction, the startup faces a fiercely competitive landscape dominated by data heavyweights. Platforms like Databricks and Snowflake are aggressively expanding their native generative AI capabilities. Snowflake Cortex now offers fully managed LLMs directly within its Data Cloud, while Databricks—which absorbed Hermann's previous venture—is building out unified platforms that merge data warehousing with conversational AI interfaces.

The existential question for independent context engines is whether they can outrun the bundled semantic layers being developed by these cloud giants. The defense strategy relies heavily on specialized, agentic workflows rather than simple Retrieval-Augmented Generation (RAG). Standard RAG architectures often depend on static knowledge bases that quickly fall out of sync with shifting business realities. The Context Factory, by contrast, continuously reads from transformation frameworks like dbt, existing dashboards, and documentation to dynamically update its understanding.

Traditional semantic layers, such as dbt or Cube, require data teams to manually define models and metrics in code. While these tools provide a robust foundation, maintaining the broader business context—the undocumented nuances of how sales territories are drawn or why a specific revenue formula changed last quarter—remains a highly manual, error-prone endeavor. The new agentic system is designed to sit on top of these existing frameworks, autonomously generating the connective tissue that makes those metrics comprehensible to a marketing manager or a chief financial officer.

Furthermore, the platform's insistence on a human-in-the-loop approval process for consequential metric changes directly addresses the enterprise fear of AI hallucination. By storing approved definitions in a versioned, customer-owned knowledge base, the system provides a verifiable audit trail. This approach transforms data engineers from reactive ticket-takers into proactive editors of corporate knowledge.

Venture analysts observing the modern data stack suggest that while consolidation is inevitable, specialized tools that seamlessly bridge the gap between raw cloud storage and business-user trust remain highly defensible. Major cloud BI tools are also rushing to integrate natural language querying. However, the prevailing industry critique is that conversational interfaces layered over poorly contextualized data simply generate wrong answers faster. By automating the tedious extraction of institutional knowledge that generic platforms struggle to capture, this new wave of agentic infrastructure aims to become an indispensable middleware, immune to the gravitational pull of any single data warehouse vendor.