Nomograph Labs

We build tools that help AI understand engineering models, then measure whether those tools actually work.

Projects

sysml

CLI and MCP server for SysML v2. Structural retrieval, graph traversal, and completeness checking for AI on systems models.

14 commands · 123 tests

nomograph/sysml-bench

sysml-bench

Benchmark harness for AI on SysML v2 tasks. Reproducible evaluation across models, tool configurations, and corpus scales.

132 tasks · 5 models · 14 findings

nomograph/tree-sitter-sysml

tree-sitter-sysml

Tree-sitter grammar for SysML v2. The parsing foundation for all Nomograph tooling.

192 tests · 89% external file coverage

What we measure

We measure AI performance on formal modeling language tasks across tool configurations, models, and corpus scales. Two findings from 132 tasks across 5 models:

FIG 1

On explanation tasks, pre-rendered model views score 0.868 versus 0.399 for agentic assembly — a 47 percentage point gap across N=3 replications.

FIG 2

Tool effectiveness is task-dependent, not tool-count-dependent. Aggregate improvement is near zero (p=0.391), but per-task effects are real and systematic. Which tools you give the model matters more than how many.

Full results at results.nomograph.ai (forthcoming).

Interested?

There are more formal languages than one group can cover. If you work with engineering models and want to understand how AI performs on them, or if you just find this kind of measurement interesting, we'd like to talk.

gitlab.com/nomograph →