1 s → 390 s
1 second of physical behavior takes 390 seconds to simulate. Runs forever on transient states. No fast iteration possible.
Early access · Montreal, QC
Constitutional AI for Hardware Engineering Teams.
An AI-governed hardware/software co-design framework that brings software-level iteration speeds to physical engineering. Built for system integrators — physics first, with a human in the loop.
21days to hardware
01engineer
80%less time debugging
§ 01 · The problem04 / phases
"Hardware is hard."
— Everyone
Legacy waterfall approaches capture failures at the absolute end of the design cycle. Every iteration costs millions of dollars and months of delay.
up to 7 yrs & $8M
ASIC: 1–3 years & $2–5M sunk. MEMS: 5–7 years & $5–8M sunk.
+4–6 months
RF antenna revisions add 4–6 months to the timeline in half of all projects.
3–6 months per update
Algorithm upgrades require 3–6 months chasing paper trails and tribal knowledge.
That's why only 6–8% of startups build hardware — while 80–85% stay in software, where every iteration is cheap.
§ 02 · The landscape03 / categories
Every engineer now runs a private, ungoverned workflow.
Every incumbent category solves a partial problem. None reconcile physics, collaboration, and AI under one roof.
Ansys · Comsol · Cadence
Traditional EDA
- ✕$320K avg enterprise license + $200/hr consulting. Inaccessible for most teams.
- ✕Siloed toolsets — high friction by design. No shared state across functions.
- ✕Zero CI/CD or hardware-software integration. Every handoff is manual.
Jira · Confluence · Slack
Siloed PLM / Dev Tools
- ✕Physics grounding: zero. Built strictly for software workflows.
- ✕Physical reality and patent tracking are not part of their design philosophy.
- ✕No traceable link from algorithm decision to hardware parameter. Tribal knowledge rules.
Claude · ChatGPT · Copilot
Generic AI
- ✕Hallucination risk: severe for BOM and circuit design. No physical constraints.
- ✕Cloud endpoints expose sensitive engineering IP with every query.
- ✕No auditability, no Physics First enforcement. Cannot be used for regulated hardware decisions.
Position
Crucible is not a build tool or generic AI. It is an engineering governance platform.
§ 03 · Why now03 / trends
Agentic AI unlocks it. Hardware now demands it.
Agentic breakthroughs make the platform buildable for the first time. Multi-sensor humanoids, edge AI, and Space 2.0 make it non-negotiable.
01
Agentic AI is here
What was PhD-grade automation last year is a commit today. Physics-Informed Neural Networks, agentic orchestration, and generated documentation finally make the platform buildable.
02
Systems are exploding in complexity
Multi-node edge AI. Humanoids carrying 60–70 sensors. Hardware–AI–software co-design is now standard. Waterfall cannot keep pace with the systems being built on top of it.
03
High-stake hardware can't wait
Space 2.0, defense, and medical demand fast cadence with airtight compliance. Paper trails decide who ships.
Crucible ships the result: fast iteration, unified agile flow, and an auditable paper trail — newly possible, newly required.
§ 04 · The framework02 / articles
Two constitutional articles. No exceptions.
Crucible is a governance platform before it is a tool. Every decision made inside it is bound by two rules — and nothing else.
Article I
Physics first
...
- Rule
- No parameter may be defined unless it traces to a first-order physically measurable quantity.
- Impact
- Guarantees traceability and explainability of AI-assistive design decisions.
Article II
Human in the loop
HIL
- Rule
- Agents act as physics probes. Humans hold ultimate decisional power based on physical evidence.
- Impact
- Guarantees alignment of AI-assistive design decisions and project purpose.
§ 05 · The pipeline05 / stages
A traceable path from idea to deployed device.
Every change that touches algorithm, firmware, or hardware generates a traceable Bill — the physical-world equivalent of commit and push in Git.
- 00
Toolchain lock & smoke tests
Toolchain failures caught here cost 20 minutes, not days.↓ - 01
Simulation
Physics model of the device domain.↓ - 02
Firmware on dev kit
Moving code onto physical components.↓ - 03
Field test
Real measurement data replays back to Stage 01 — simulation evolves with evidence.Replay ↺ - 04
Host integration
Smart home, gateway, cloud deployment.↓
Field measurements feed back into simulation — the digital twin is a byproduct of every developed device.
§ 06 · The engine03 / modules
Three fundamental pillars under one governance layer.
Physics into code, debate into documentation, local compute into a privacy-preserving agent swarm.
01
90% less tokens used
Local-cloud hybrid
Local agents handle sensitive code, JSON, UART, and raw data. Code and data never leave the workstation. Cloud handles unidirectional research only.
02
−80% less debug time
Judicial-style decisions
AI agents debate design options, surfacing evidence and challenging assumptions. Reduces debug evidence collection time by 80%. Decisions are automatically documented.
03
Weeks → 1 day
Auto PINN generation
AI agents automatically generate Physics Informed Neural Networks, transforming physics equations into signals. Reduces simulation tuning from weeks to a single day.
§ 07 · Case study01 / device
GaitSense — 21 days, one device.
A single-ankle wearable that detects walking asymmetry. Built entirely inside Crucible, with agent-assisted debugging before any hardware was fabricated.
MCU
nRF52840
Sensor
LSM6DS3TR-C
Radio
BLE
Form
Ankle wearable
Legacy approach
3–6months
- —Expected timeline: 3–6 months.
- —Team of several engineers. Software, Firmware, Electronics, Algorithm.
- —High risk of late-stage algorithm failure due to simulation-to-reality disparity.
- —Months spent compiling compliance documentation.
Crucible accelerated
21.days
- ●Timeline: 21 days, end to end.
- ●One single engineer.
- ●Agent assistance caught a stance-imbalance hallucination while testing stair-walking logic — a bug that usually requires $1,000 sensor-fusion wearables to surface.
- ●Compliance documentation squeezed from months into days, ready for patent filing.
§ 08 · The roadmap
Expanding the platform across physical domains.
Crucible begins with hardware-dense startups and widens outward — toward regulated, deep-tech, and extreme-environment engineering.
Phase 01
Domain expansion
- → Crucible-ComfortAdvanced HVAC integrations
- → Crucible-ConnectTelecommunications
Phase 02
Deep tech & AI
- → Crucible-NeuronFormalized automatic PINN generation
- → Crucible-HybridStandalone hybrid-intelligence infrastructure
Phase 03
The future
- → Crucible-SiliconASIC and MEMS
- → Crucible-SentinelDefense applications
- → Crucible-CompanionRobotics
- → Crucible-BeyondCommercial space
§ 09 · The team02 / founders
Built by physicists and serial deeptech founders.
Two PhDs. Two prior deeptech startups. One platform we wish had existed.
Siyao Shao
Co-founder · Technical Lead
Serial entrepreneur steeped in the US and Canadian startup environments. Founded two deeptech companies prior.
Expert in physics-native computing and AI system design.
Roxanne Turcotte
Co-founder · Scientific Lead
Astroparticle physicist from IceCube and SNOLAB. Saw bad hardware quietly ruin good analyses.
Signal analysis, multi-physics simulations, and precision-detector work.
§ 10 · Engage
Hire our team to build your hardware.
While the platform is maturing, we take on a limited number of hardware-development engagements each quarter. Two PhDs and a deeptech-native team, applied directly to your device.
01
Proof of concept
per project
Physics-first feasibility studies for embedded, wearable, and HVAC programs. Delivered with the simulation, firmware, and compliance artifacts to defend the design.
02
Development partnership
team setup + engagement
End-to-end hardware–firmware builds under the Crucible governance model. We embed alongside your team through Stages 0–4 — from toolchain lock to field deployment.
03
Advisory & audit
retainer or fixed
Diagnose stalled hardware programs, audit simulation-to-reality gaps, or stress-test a BOM and firmware stack before you commit to fabrication.
Active domainsWearables & BLE·HVAC & sensor fusion·RF & telecom·Smart home & IoT
Request a consultation
Have our team develop for you.
Direct line
info@cruciblestudio.caOpen email→Early access
Ship hardware like software.
We're working with early design partners in wearables, HVAC sensors, telecom, and robotics. If you ship physical systems and you're tired of the waterfall of death — let's talk.