Controlled systems studio

From uncertain instruction to controlled outcome.

NCM Labs builds software that reaches across the boundary between reasoning and effect—from typed agent systems to CAD automation, CNC tooling, and machine firmware.

Talk to us Explore the work
Agent systems CAD & fabrication Machine control

LIVE SYSTEM MAP

Supervised agent flow

TRACE ACTIVE
Instruction-to-outcome system map A four-stage flow moves from an instruction through an uncertain boundary and validation gate to controlled execution. INSTRUCTION typed task ORACLE uncertain<T> BOUNDARY validate OUTCOME traced action THE STOCHASTIC BOUNDARY IS VISIBLE confidence · budget · escalation

Intent is reasoned about, checked, then allowed to affect the system.

Public work / current signal

The work makes the position concrete.

Four public projects. One throughline: make the path from instruction to effect explicit, testable, and useful.

02 AI-to-CAD bridge · Python

Fusion 360 MCP

An MCP server that lets an AI assistant inspect and manipulate a live Autodesk Fusion 360 session—from sketches and features to measurements, joints, and validation.

Conversation becomes inspectable geometry.

PythonMIT
View repository
03 CAD-to-machine tooling · nesC

Carvera Processors

FlatCAM post-processors for the Makera Carvera Air, producing milling, drilling, and laser G-code in the Smoothieware dialect.

Toolpaths become machine-ready instructions.

nesCMIT
View repository
04 Machine control · C++

Makerfarm i3v

A Marlin 2.1 firmware configuration for the Makerfarm i3v 3D printer, bringing modern motion control to a proven physical platform.

Validated firmware becomes precise movement.

C++GPL-3.0
View repository

Capabilities / where we contribute

Research depth. Delivery discipline.

We collaborate where software has to reason, cross a boundary, and produce an outcome that matters.

AGENT.SYSTEMS

AI & agent systems

We design agent runtimes where uncertainty, budgets, supervision, and escalation paths are visible before they become production incidents.

FORGE LLM orchestration supervision evaluation

PHYSICAL.COMPUTE

CAD & fabrication tooling

We connect reasoning systems to CAD, toolpath, and machine workflows with explicit validation between intent and action.

Fusion 360 MCP CNC G-code

MACHINE.CONTROL

Firmware & embedded

We work at the layer where software becomes motion: firmware configuration, device constraints, and dependable machine behavior.

Marlin C++ motion control hardware

DELIVERY.LAYER

Software & data engineering

We build the reliable delivery layer behind ambitious systems—from real-time Phoenix applications to Go services and event-driven data pipelines.

Elixir Phoenix Go event systems

Approach / four invariants

Control is an architectural property.

A polished demo is easy. A system that behaves honestly under uncertainty, failure, and consequence takes a different set of decisions.

boundary.forge 
result = reason prompt
when result.sure(above: 0.85) ->
  give validated(result)
else ->
  escalate to human
  1. 01

    Name the uncertain boundary

    Stochastic work should be visible in the architecture, not disguised as a normal function call.

  2. 02

    Validate before effect

    Reasoning can explore. A typed or physical action must pass an explicit boundary first.

  3. 03

    Keep a human ceiling

    When confidence or consequence demands it, the system escalates instead of improvising.

  4. 04

    Trace what actually happened

    Decisions, costs, fallbacks, and execution paths should leave evidence an operator can inspect.

Open channel / start a conversation

Working on a system where the boundary matters?

Tell us what needs to reason, what needs to stay deterministic, and what the outcome has to control.

[email protected]
Available for focused technical collaborations

No CRM. No mailing list. Your message is sent directly and not stored here.