One Question, Five Plausible Answers: dv01's Semantic Layer Gives AI Context

Ask a general-purpose LLM to calculate cumulative net loss, and it will likely produce a credible formula: charge-offs less recoveries, divided by a pool balance. But following a formula is not the same as producing the right analysis. 

dv01’s Model Context Protocol (MCP) addresses that gap by connecting large language models to dv01’s semantic layer and analytical infrastructure. Rather than relying on a model to interpret a methodology on its own, MCP gives it access to the governed data definitions and dv01 capabilities required to apply the appropriate methodology to the underlying data.

Why context matters in collateral analysis

At first glance, CNL appears straightforward: charge-offs less recoveries, divided by a pool balance. But each part of that calculation can involve important methodological decisions.

For a closed, static pool, the denominator may be the purchase balance established at closing, allowing losses to be measured consistently as the pool pays down. For broader market or vintage analysis, origination balance may be more appropriate.

The numerator can also require interpretation. Issuers and servicers may classify charge-offs, recoveries, bankruptcies, and other loan events differently. Standard reporting fields may not capture the full economics of every transaction, requiring documented adjustments for specific deals or servicers. The resulting calculation may be mathematically correct while still answering the wrong question.

The same request can produce very different results depending on the data, definitions, and methodology available to the model.

The dv01 Semantic Layer turns plausible answers into grounded analysis

Rather than asking AI to learn a new formula, connecting an LLM to dv01’s agent-ready infrastructure gives it the data definitions, methodologies, and context needed to apply that formula correctly. Analysts should not have to reconstruct institutional knowledge every time they ask a question.

dv01’s semantic Layer gives standardized loan-level data consistent business meaning. It maps source fields to defined concepts, establishes how metrics should be calculated, and incorporates documented deal- and servicer-specific exceptions. This creates a shared analytical foundation that can apply the same defined logic across workflows.

Through MCP, AI tools can call dv01 capabilities built on this foundation rather than attempting to interpret the data or recreate the methodology inside the model. The result is analysis grounded in defined data and methodologies, with greater consistency and transparency.

Bring dv01 into your AI workflows

dv01 is building the agentic infrastructure for structured finance, giving firms multiple ways to move AI from experimentation into everyday work. With dv01’s MCP, teams can connect the AI tools they already use to standardized loan-level data and specialized analytical capabilities. Built-in Assistants for Credit Facility Management and DealStudio bring AI directly into facility management and deal structuring workflows within dv01.

Whether firms want to extend their existing AI environment or use purpose-built assistance within the dv01 platform, dv01 provides a practical path to operationalize AI across their workflows.

Contact us to explore which approach is right for your firm.