Projects

Things I have built and worked on, in and out of working hours.

MBE3Dstudio

A 3D model-based engineering studio: geometry, allocation, requirements, and executable constraints in one versioned model. It is the product side of Vector Stream Systems LLC, and the subject of a pending U.S. provisional patent application.

The core problem it addresses is one I hit repeatedly on aircraft and defense programs: requirements, architecture models, and verification evidence live in separate tools with no enforced relationship between them, so they drift. By the time a review board notices, months of design work are committed.

MBE3Dstudio connects those tools into a single graph with formal traceability, hybrid retrieval over that graph, and change propagation. When an interface definition changes, the system identifies every downstream requirement that needs review before the change lands rather than after.

Technically it is a hybrid: an ontology layer for the parts of the domain that benefit from formal semantics, and a plain directed graph for the parts that benefit from being fast and inspectable. It is built in Rust and React, with CLI and API surfaces, and exposes context to AI tooling through the Model Context Protocol so that agents read structured, versioned system state through a controlled interface instead of scraping documents.

MBE3D MCP server

A Model Context Protocol server that gives AI assistants governed access to an MBSE model. Rather than pointing a language model at a pile of exported documents, it exposes typed operations, query requirements, traverse traceability relationships, check constraint satisfaction, retrieve related design context, so that what the model sees is scoped, current, and attributable.

The design principle is that an agent should not be able to assert something about the system that the model cannot back up. Retrieval is grounded in the graph, and validation runs against rules the agent does not control.

Model-based safety and hazard assessment tooling

At Boeing I performed user acceptance testing on tooling that generated safety and hazard assessment artifacts directly from system models for certification-oriented analysis. The value was not the automation itself but the coupling: when the architecture model changed, the hazard analysis was no longer silently stale.