AI-native engineering for the physical world
From idea to
physical reality.
Terranoux is building the intelligence layer between a human idea and a validated, manufacturable physical product.
- Stage
- Early company · research and architecture
- First focus
- Mechanical components
- Long-term scope
- Requirements to manufacturing
- On this page
- Illustrative demonstrations, labelled
01Demonstration · illustrative
One sentence in. An engineered, buildable vehicle out.
Walk the autonomous inspection vehicle from intent through requirements, trade-offs, drawings, materials, simulation assumptions and manufacturing.
Intent · natural language
Scripted demonstration · not a live system
“A portable inspection platform for rough terrain that carries a 3 kg sensor package through a full 8 km survey, works from −10 to 45 °C in rain, and one person can carry. Budget about $6,000.”
Intent becomes requirements, traced to their words.
Hover or tap a requirement to see the words it came from. Three of six are interpretations, and the system should say so rather than guess silently.
Every value, drawing and decision in this panel is hand-written to illustrate the reasoning. No geometry is generated, nothing is simulated and nothing has been built.
What exists today ↓02Connected constraints
Change one requirement. Watch the rest of the design respond.
Engineering decisions are coupled. More range needs a bigger battery, which adds mass, which needs more energy per kilometre. Move the sliders and see which requirements still close.
Requirements
Consequences
Simplified illustrative model
Vehicle mass
≤ 18 kg · one-person carry
16.9 kgpass
BOM cost
≤ $6,000
$5,400pass
Frame safety factor
≥ 2.0
2.4pass
- Battery
- 0.63 kWh
- Battery mass
- 4.2 kg
- Usable at min. temp
- 76%
- Frame mass
- 3.1 kg
What propagated
Change a requirement to see how it travels through the design.
+− Model and constants used
frame mass = ρ × 1148 cm³ × (1 + 0.06 × (payload − 3))
usable capacity = 1 − 0.008 × (20 − T_min)
energy = range × 0.05 kWh/km × (vehicle + payload)/20 kg × 1.2 ÷ usable
battery mass = energy ÷ 0.15 kWh/kg · vehicle = 9.6 kg + frame + battery
cost = $4,650 + $650/kWh + frame material (+$420 if not weldable)
safety factor scales with yield strength and load, from 2.4 at baseline
First-order relationships with stated constants, chosen to reproduce the example. Material properties are nominal handbook values. Not an engineering calculation and not Terranoux output.
03The engineering lifecycle
Understand → Design → Model → Simulate → Experiment → Prototype → Validate → Manufacture.
Eight stages from intent to production. The loop runs backward whenever evidence disproves an assumption, which is most of engineering.
Understand
Turn intent into requirements and constraints.
What must the thing do, in what environment, for how long, at what cost? Vague intent becomes measurable requirements, and the requirements that conflict are surfaced early.
- requirements
- environment
- cost targets
- standards
output → Requirements with acceptance criteria
The loop runs backward whenever evidence disproves an assumption. Pick a case.
04Feasibility and validation
A design is only an idea until it can be made, and reality agrees.
Manufacturing feasibility
IllustrativeCheck the part against the process before anyone cuts metal.
Ø22 H7 bore, laser cut ±0.10 mm
process cannot hold the fit → add CNC finish pass
Bend radius 2 mm on 3 mm 6061-T6
below 1.5 t → crack risk, raise to 4.5 mm
Weld at bracket root
in peak stress zone → move weld 15 mm outboard
M5 tapped holes, 8 mm deep
standard tooling, no change
DRV-01 mounting bracket, item 4 in Fig. 01.
Physical validation
IllustrativeWhen the measurement disagrees, the model changes.
- 1Hypothesis
- 2Experiment
- 3Measurement
- 4Design update
Inspection vehicle example. Track losses and thermal margin were worse than simulated; battery and heatsink were revised. No such test has been run.
05Today and the roadmap
Start with mechanical components. Build toward the whole lifecycle.
The ambition is broad. The first proof should be narrow, measurable and honest. Mechanical parts are where an engineering system can be checked against reality fastest.
Why mechanical first
Recommended first domainMature physics
Structural and thermal analysis of parts is well understood, so errors are attributable.
Cheap to verify
A load test and a dimensional inspection settle whether the prediction held.
Accessible manufacturing
Machining, sheet metal and printing are available at prototype quantities in days.
Clear success criteria
A part meets its requirements, fits its assembly and can be made at cost, or it does not.
Brackets, housings, mounts, shafts and enclosures first. Electrical, materials, manufacturing and robotics remain in scope.
- 01intent
Idea
- 02loads · limits
Specification
- 03geometry · datums
Engineering model
- 04stress field
Simulation
- 05printed iteration
Prototype
- 06machined part
Physical object
What exists, and what does not
available in research planned- Draft formatsExperiment contract, risk envelope and evidence package, published on this site.Published draft
- Interactive demonstrationsThe scripted examples on this page. They illustrate reasoning; they do not compute it.Illustrative
- Connected engineering representationRequirements, decisions, geometry, analysis and cost linked in one structure.In research
- Mechanical component pipelineRequirements → CAD → FEA → manufacturability for single parts. The first validation domain.In research · first focus
- Physical experiments through adaptersBounded tests on existing equipment, returning raw evidence.Planned
- Governed procurement and ordersPurchases and manufacturing orders proposed by Terranoux, approved by LucidRail.Planned
- Multidisciplinary assembliesMechanical, electrical, materials and controls in one loop.Long term
Not offered today: no customer product, no laboratories or factories, no equipment adapters and no generated geometry. Details on the research page.
06Three systems
Aletheonix discovers. LucidRail governs. Terranoux builds.
Three separate systems, each useful on its own. Terranoux works from human ideas directly, and can take work from Aletheonix. Consequential actions need LucidRail's approval, which Terranoux cannot grant itself.
- 01Discover
What should we understand?
Notices what is uncertain or failing, forms hypotheses and weighs evidence.
- 02Govern
What may AI do?
Identity, permissions, budgets, approvals, policy and audit for AI actions.
- 03Build
Terranoux
How do we make it real?
Engineers ideas into designs, simulations, experiments, prototypes and products.
01 · Terranoux
Proposes an engineering action
- Purchase parts
- Operate equipment
- Run experiments
- Modify machines
- Submit manufacturing orders
02 · LucidRail
Evaluates authority, budget and risk
Identity, permissions, spending limits, policy and required approvals. Outside Terranoux, by design.
03 · Approved action executes
Inside the limits that were granted
Or it does not execute, and the refusal is recorded with its reason.
07Work with us
Looking for engineering research collaborators and pilot partners.
Research collaborators
Researchers in computational design, simulation, materials, manufacturing and verification of AI decisions.
- Open problems listed plainly on the research page
- Shared benchmarks for idea-to-part engineering
- Honest evaluation against physical measurement
Pilot partners
Teams that design and make mechanical components, and can test what they build.
- Real parts with real requirements, not toy problems
- Access to measurement: load tests, inspection, field data
- Willing to tell us where the approach breaks
Terranoux is building the intelligence layer between an idea and the physical thing it becomes.
Or write directly: benjamin@lucidrail.com