About
Where AI leverage lives in structured, long-running work, found with open tools and reproducible measurement.
Most AI tooling optimizes for the single turn. We work the harder problem: multi-session work across large codebases, formal engineering artifacts, and collaborative projects where the context outlives any one conversation.
Context engineering
How you present information to a model matters more than how you retrieve it. We build tools that make structured artifacts legible to AI through composable CLI commands and MCP servers.
Spec management
Long-running work needs structure that persists across sessions. Synthesist tracks specs, task dependencies, and propagation chains. The human and the agent use the same interface.
Reproducible evaluation
Claims about AI performance need measurement. Our benchmarks are built for independent replication: open tasks, open data, structured scoring, statistical frameworks.
How you present information to a model matters more than how you retrieve it.
Representation interventions, like pre-rendering a model view for the task at hand, produced large and replicable effects. Retrieval interventions, like vector search and graph traversal, produced null results. It is a consistent shape across our benchmarks, and the through-line of the work.
How this started
This began as an academic exploration of how AI interacts with structured engineering artifacts. We built a tree-sitter grammar for SysML v2, then a CLI tool, then a benchmark harness to measure whether the tool actually helped. Each step produced something others could use and something we could measure.
The method is to build evaluation apparatus and look for where high-leverage AI use can be found. Model-based systems engineering is one area we think has real potential: the industries that lean on MBSE, like defense, aerospace, automotive, and medical devices, tend to adopt modern AI tooling slowly. We would like to help them get there through open instrumentation, and to grow a community around measured, LLM-mediated engineering.
Papers
Papers are in progress, and each repo opens with its release. Representation Over Retrieval, on how presentation beats retrieval for AI on structured engineering tasks, was submitted to GVSETS 2026.
The name
A nomograph is a graphical calculation instrument: three parallel scales, each graduated differently. An isopleth, a line laid across the scales, reads the answer by alignment. No computation. The structure of the instrument encodes the equation.
Our mark is an isopleth crossing three scales. Each scale has its own graduation rhythm. The curve reads from top-left to bottom-right. The N emerges from the reading.