Controlled systems studio
NCM Labs builds controlled intelligent systems across the full stack—from typed agent systems and model evaluation to CAD automation and embedded research.
LIVE SYSTEM MAP
Supervised agent flow
Selected work / current research
Two public projects and three private R&D threads. One throughline: make the path from instruction to effect explicit, testable, and useful.
A programming language for oracle-augmented computation. FORGE makes uncertainty, supervision, cost, and deterministic boundaries part of the program instead of afterthoughts around an LLM call.
Uncertain results become explicit, handled values.
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.
We build reproducible loops for adapting models to specific tasks, treating data curation, training, serving, and held-out evaluation as one system.
Capability changes are measured against explicit baselines.
We are exploring bounded worker systems that plan, execute, critique, and improve work against explicit acceptance criteria, with evidence preserved across every decision.
Autonomy advances only when evidence clears the gate.
Ongoing research into sensor-rich devices and distributed control, with firmware, data collection, feedback, and safety designed together from the start.
Intelligence meets the physical world through controlled interfaces.
Capabilities / where we contribute
We collaborate where software has to reason, cross a boundary, and produce an outcome that matters.
AGENT.SYSTEMS
We design languages and agent runtimes where uncertainty, budgets, supervision, and escalation paths remain visible from intent to outcome.
MODEL.EVALUATION
We treat data quality, adaptation, serving, and held-out evaluation as a single evidence loop rather than disconnected experiments.
CAD.AUTOMATION
We connect reasoning systems to live design environments with inspectable operations and explicit validation between intent and geometry.
EMBEDDED.INTELLIGENCE
We explore how sensing, firmware, feedback, and safety constraints become dependable behavior in physical devices.
Approach / four invariants
A polished demo is easy. A system that behaves honestly under uncertainty, failure, and consequence takes a different set of decisions.
result = reason prompt
when result.sure(above: 0.85) ->
give validated(result)
else ->
escalate to humanStochastic work should be visible in the architecture, not disguised as a normal function call.
Reasoning can explore. A typed or physical action must pass an explicit boundary first.
When confidence or consequence demands it, the system escalates instead of improvising.
Decisions, costs, fallbacks, and execution paths should leave evidence an operator can inspect.
Open channel / start a conversation
Tell us what needs to reason, what needs to stay deterministic, and what the outcome has to control.
[email protected]We can't find the internet
Attempting to reconnect
Something went wrong!
Attempting to reconnect