01
The AI layer
The hard part was the substrate, not the model
AI on transportation data is only as good as the data underneath it. The difficulty was never the model. It was getting every agency's legacy feed into one normalized, queryable shape. Once that substrate exists, grounded AI access follows directly.
02
MCP
Native to every node and hub
Agents query the same normalized warehouse an analyst does, through the same primitives. No custom integration, and nothing leaves your infrastructure.
- Same warehouse. Claude, GPT, and local models read the system of record directly, not a copy.
- Same primitives. The endpoint exposes the node's own query surface, so agent answers and analyst dashboards agree.
- No new integration. MCP ships with every node and hub. Point an agent at it.
03
Boundaries
Grounded and single-tenant
A copilot sits over your dashboards and answers natural-language questions across queries at once. Answers come from your system of record through MCP, not from the model's own guess. Prompts and data never cross customer boundaries.