Engineering Simulation Platform · Now in Early Access

Engineering Simulation
in Minutes, Not Days

Upload your CAD. Describe your problem. Get stress, displacement, and safety factor results in minutes, with ML previews in under a second — no meshing, no solver config, no specialist required.

The old way
Export CAD → Format Convert → Import → Clean Geometry → Mesh → Materials → Setup BCs → Solver Config → Run → Post-process
phinen simplifies this
With phinen
Upload
Describe
Results
Sub-second ML inference Massive cost reduction Physics, simplified
Try with a Sample Part → See How It Works
< 0.8s
GNN Inference
Sub-second stress & displacement on any geometry
4
Solver Tiers
GNN → PINN → Hybrid → FEniCSx. Start fast, escalate for proof.
6
Step Guided Wizard
Upload → Material → Select → Chat → Plan → Run. Zero setup expertise.
0
Solver Config Required
AI writes the entire simulation plan from your plain-English description.
Live Platform Preview

AI-Guided
Structural Simulation Wizard

From STL upload to von Mises stress map in minutes, with ML previews in under a second. The AI plans the simulation, evaluates your boundary conditions, and interprets the results — before and after every run.

🔒 fea.phinen.com
phinen platform: a ball stud, retaining ring and socket housing in the 3D viewer, with the AI advisor reviewing the support, load and material before the simulation plan is generated

Screenshot of the live platform: the AI advisor checking a contact-analysis setup · fea.phinen.com ↗ · No signup required for the demo

See It In Action

From CAD File to Engineering Insight in 6 Steps

Upload your geometry, describe your problem in plain language, and get a physics-grounded simulation result — no meshing, no solver setup, no specialist required.

📁
Step 01
Upload STL / CAD
Drop in an STL file. The 3D viewer renders your geometry instantly — AI immediately analyses shape, dimensions, and topology.
🧪
Step 02
Pick Material
Select from a curated library (Steel, Aluminium, Titanium, PEEK and more). E, ν, ρ, σy auto-populated. AI confirms the choice fits your geometry.
🎯
Step 03
Tag Selections
Click faces in the 3D viewer to mark fixed supports, load regions, symmetry planes, and refinement zones. AI suggests placements based on geometry analysis.
💬
Step 04
Chat with AI Advisor
Ask "Are my selections correct?" and the AI evaluates each one — flags under-constrained setups, wrong load directions, or unrealistic magnitudes before you run.
⚡
Step 05
Generate Plan & Run
One click generates a full simulation plan from your inputs. Choose GNN for instant results or escalate to PINN/FEniCSx for higher-fidelity validation.
📊
Step 06
Results & Insight
Stress, displacement, safety factor fields rendered on your geometry. AI interprets failure zones, suggests design changes, and compares against material limits.
🔒 fea.phinen.com
phinen walkthrough, step 1: upload

Drop an STL into the workspace to start a new simulation.

Try It on Your Geometry →
From Geometry to Results — How phinen Works

Walkthrough of the full simulation pipeline · Use ◀ ▶ to navigate · Space to pause

Solver Intelligence

Start Fast. Escalate for Proof.

Four solver tiers — each built for a different stage of the design cycle. Choose exactly how much physics fidelity you need at each stage.

Fastest
🕸️
GNN Preview
MeshGraphNet
< 0.8s
inference time
Graph neural network runs directly on your 3D mesh. Instant stress and displacement fields for rapid design exploration and early-stage iteration.
Speed●●●●●
Accuracy●●●○○
Use forConcept exploration
🔢
PINN Preview
Physics-Informed NN
~2s
inference time
Conservation laws (mass, momentum, energy) embedded directly into the neural network. Physics-constrained predictions that respect governing equations.
Speed●●●●○
Accuracy●●●●○
Use forDesign validation
⚡
Hybrid Preview
GNN + PINN Corrector
~3s
inference time
GNN produces the fast initial field; a PINN corrector layer refines it against physical constraints. Best accuracy-to-speed trade-off for pre-certification.
Speed●●●●○
Accuracy●●●●○
Use forPre-certification
Gold Standard
🔬
FEniCSx Result
Full Finite Element Solve
~2 min
full FEA solve time
Full PDE solve on the auto-generated mesh. Exact boundary conditions, nonlinear materials, certifiable results. Used for final design sign-off and surrogate retraining.
Speed●●○○○
Accuracy●●●●●
Use forCertification / sign-off

All four solvers are available in a single click from the solver selector. phinen automatically suggests the right tier based on your design stage and required confidence level.

AI Advisor

An AI That Checks Your Work
Before You Run

A built-in AI advisor evaluates your boundary conditions, load placements, and selection logic — and gives you a clear GO / NO-GO before wasting compute.

⚠️
Catches Setup Mistakes Early
Under-constrained (no fixed support), over-constrained (symmetry conflict), or loads pointing in the wrong direction — flagged before the solver runs.
📐
Validates Positions Against Geometry
The AI reads the actual 3D coordinates and radii of your selections and checks they correspond to structurally meaningful regions of the part.
🔢
Suggests Realistic Load Magnitudes
Based on the part dimensions, material, and geometry type, the AI recommends realistic force (N) or pressure (MPa) ranges — so you don't start with a 10× wrong order of magnitude.
✅
Confirms When You're Ready to Run
Once fixed supports, loads, material, and magnitudes all check out, the AI tells you clearly: "Setup looks good — generate the plan." No guessing.
AI Advisor · Pre-Run Check Step 4 of 6
👤
Are my selections good for this bracket analysis?
π
✓ PARTIAL — 2 of 3 checks pass
✓Fixed support at (0, 60, 105) mm — correct for a cantilevered bracket.
⚠Load direction (0, 0, -1) — pointing in −Z. Confirm this is vertical gravity loading, not axial.
✗No load magnitude set — for a 62 mm steel bracket I suggest 500–2,000 N.
Set the magnitude and confirm the direction — then you're ready to generate the plan.
👤
Yes, gravity. Set 1,000 N.
π
✓ Setup looks good — fixed support, 1,000 N gravity load, Steel 304. Click Generate Plan to proceed.
Under the Hood

Engineered for Speed, Accuracy & Physical Credibility

Neural-network solvers compute the physics, with a classical finite element solver to verify them. An LLM handles only the conversation: it turns your request into a simulation setup and explains the computed results. Covers structural, thermal, and fluid simulation.

1
You describe
Upload a part and describe the load case in plain language.
2
LLM plans
An LLM turns the request into a structured setup: material, loads, supports, solver.
3
Neural networks solve
MeshGraphNet, PINNs and CANN/TANN compute the stress, displacement, thermal and flow fields on RunPod GPUs. FEniCSx steps in when confidence is low.
4
LLM explains
Reads the computed results and writes the engineering report: findings, risk areas, recommendations.
🟢
NVIDIA PhysicsNeMo
Physics-ML Framework
The framework phinen's neural solvers are built on. We use PhysicsNeMo's MeshGraphNet architecture and physics-ML tooling to train, fine-tune and serve our surrogate models for engineering simulation.
Core ML Platform
🕸️
MeshGraphNet
Graph Neural Network Solver
Represents the part's mesh as a graph: nodes hold local physical state, edges carry interactions between neighbours. Predicts full displacement and stress fields across complex 3D geometry in about 0.3 s, with a confidence score on every result.
Graph Neural NetSub-second Inference
🔢
Physics-Informed Neural Networks
Physics-Constrained Solver
Neural networks trained directly against the governing equations for elasticity, heat transfer and fluid flow. They run on their own, or as a corrector that refines MeshGraphNet output against physical laws (Hybrid mode).
PINNPhysics Constraints
🧪
CANN & TANN
Neural Material Models
Learned constitutive models that capture how real materials behave beyond the linear range. CANN covers hyperelastic materials such as rubber and elastomers; TANN covers metals past yield. phinen picks the right model from the material's properties.
Constitutive AINonlinear Materials
🔬
FEniCSx
High-Fidelity Validation Solver
When the confidence gate flags a low-certainty neural result, FEniCSx automatically runs a full finite element solve. Results are validated, marked Gold tier, and fed back to improve the surrogate. Also used for full-fidelity and contact analysis.
FEM FallbackGold Validation
🧠
Large Language Model
Language Layer
Handles the conversation, not the physics. It turns plain-language requests into a structured simulation setup, advises on geometry and parameters, and writes the engineering report from the solvers' results. Stress, displacement and safety-factor values always come from the solvers.
LLM PlanningResult Reporting
⚡
RunPod GPU Cloud
Training · Inference
phinen's neural networks are trained and run on RunPod. A dedicated GPU pod trains the models (MeshGraphNet, PINNs, CANN, TANN) and hosts the inference server that runs every simulation. A separate pod runs the web app and API and connects securely to the GPU pod.
GPU ComputeModel TrainingInference
☁️
Cloud-Native Infrastructure
Storage · API · Streaming
Python and FastAPI backend with SSE streaming, Cloudflare R2 for geometry and result storage, and a Next.js 14 + Three.js frontend with a real-time 3D viewer. Built AI-first from the ground up for scalable collaboration and digital twin integration.
Cloud-NativeDigital Twin Ready
Built with
Built in Python (solvers, ML, backend) and TypeScript (frontend).
ML & Neural Solvers
  • PyTorch
  • NVIDIA PhysicsNeMo
  • NVIDIA Modulus
  • PyTorch Geometric
Simulation & Meshing
  • FEniCSx
  • PETSc
  • Gmsh
  • trimesh
Language Model
  • LLM for planning & reporting
Application
  • FastAPI
  • Next.js 14 · React
  • Three.js
Infrastructure
  • RunPod GPU Cloud
  • Cloudflare R2
Simulation Domains

One Platform. Multiple Physics.

phinen handles the full range of engineering simulation workloads — from structural FEA to thermal management to fluid dynamics — with digital twin integration on the roadmap.

🔩
Structural FEA
Stress · Displacement · Fatigue · Safety Factor
Run stress analysis on mechanical components and assemblies in seconds. Identify failure zones, verify safety margins, and iterate on geometry — without waiting for a full FEM solve every time.
Static AnalysisFatigue & Life
🌡️
Thermal & Fluid (CFD)
Temperature · Velocity · Pressure · Heat Transfer
Predict airflow patterns and temperature distributions across complex geometries. Detect hotspots, optimise cooling configurations, and validate thermal margins — early in the design cycle.
Thermal ManagementAirflow
⚙️
Multiphysics (Roadmap)
Coupled Thermal-Structural · Vibro-Acoustic · Fluid-Structure
Real-world problems rarely involve a single physics domain. phinen is building toward coupled simulations — thermal expansion driving structural stress, fluid pressure loading structures, and more.
Coupled SimulationFSI
🏭
Digital Twin (Roadmap)
Continuous Simulation · Live Asset Monitoring · Predictive Insight
Every simulation run builds a labeled dataset that improves the surrogate model over time. The vision: a persistent simulation layer that mirrors your physical asset — predicting behaviour before failures occur.
Self-ImprovingLive Insight

Works Across Industries

Automotive & EV Aerospace & Defense Electronics & Semiconductors Industrial Equipment Energy & Power Medical Devices Healthcare & MedTech Robotics & Automation Consumer Products Supply Chain & Logistics Construction & Infrastructure Defense & Space
Key Benefits

Why Teams Choose phinen

Directly from our platform design principles — speed, credibility, and accessibility without compromise.

⚡
Simulation Cycles: Hours/Days → Minutes
ML surrogate inference runs in under a second. End-to-end — from geometry upload to engineering report — completes in under 2 minutes, for any physics domain.
🎯
Physically Grounded Results
PINNs encode conservation of mass, momentum, and energy. The confidence gate and FEniCSx fallback ensure physical credibility is never sacrificed for speed.
👩‍💻
No Simulation Specialist Required
The guided setup interface walks any engineer through boundary conditions, loads, and material properties. No manual meshing, no solver scripting, no post-processing expertise needed.
🏭
Digital Twin Ready
Cloud-native deployment with structured data storage. Infrastructure-ready for digital twin integration — phinen is designed to be the simulation intelligence layer.
🔄
Self-Improving with Every Run
Every simulation auto-registers a labeled training sample. The more you use it, the more accurate the surrogate models become for your specific geometries.
📊
Rapid Design Iteration
Run multiple geometry variants in minutes. phinen enables rapid side-by-side comparison across structural, thermal, or fluid scenarios — accelerating confident trade-offs between cost, performance, and reliability.
Why phinen

The Tool Your Stack Was Missing

Legacy simulation tools were built before foundation models existed. phinen was designed AI-first from day one.

phinen
AI-native · all tiers
ANSYS / Abaqus
Legacy FEA
OpenFOAM
Open-source CFD
Traditional FEA
In-house workflow
Time to first result Minutes (sub-second preview) Hours – days Hours – days Days – weeks
Setup required Guided wizard Manual meshing + BCs Scripting required Specialist required
AI planning & advice ✓ Full ✗ ✗ ✗
Solver options GNN / PINN / Hybrid / FEniCSx FEM only CFD only FEM only
License cost Free → $199/mo $$$$ /seat/year Free (GPU cost) $$$$ + headcount
Geometry → result 3 clicks 50+ steps Dozens of steps Weeks of work
Confidence validation ✓ Auto confidence gate ✓ Manual Manual Manual
Pricing

Start Free. Scale as You Grow.

No credit card required during early access. All plans include structural, thermal, and fluid simulation.

Starter
Free / month
Perfect for individual engineers evaluating the platform or running occasional analyses.
  • 10 simulations / month
  • STL / STEP file upload
  • Structural FEA · Thermal · CFD
  • LLM conversational interface
  • PDF engineering report
  • Multi-variant comparison
  • API access
  • Custom surrogate training
Get Started Free →
Most Popular · Early Access
Professional
$199 / month
For engineering teams running regular simulation workflows and design iterations.
  • 200 simulations / month
  • All geometry formats
  • Structural · Thermal · Multiphysics
  • LLM conversational interface
  • PDF + JSON engineering reports
  • Multi-variant comparison (up to 5)
  • REST API access
  • Custom surrogate training
Start Free Trial →
Enterprise
Custom pricing
For organisations requiring high-volume simulation, custom model training, and dedicated infrastructure.
  • Unlimited simulations
  • All geometry formats + batch API
  • Full multiphysics suite
  • Custom surrogate model training
  • Digital twin integration
  • Multi-variant optimisation
  • Dedicated GPU infrastructure
  • SLA + priority support
Contact Us →

All plans are free during the early access programme. Pricing takes effect at general availability.

About phinen

Built for Industrial-Grade Engineering Simulation

phinen is an engineering simulation acceleration platform designed to help teams run structural, thermal, and fluid analyses on complex geometries — early, rapidly, and with high confidence.

Traditional FEA and CFD workflows require complex meshing, manual boundary setup, long solver runtimes, and specialist expertise. phinen replaces this fragmented process with a guided workflow that blends ML surrogate models, physics-informed constraints, and hybrid validation — delivering fast, interpretable results without sacrificing physical credibility.

Primary Objective

Accelerate engineering design decisions by transforming complex simulation workflows into guided, interpretable, and physically consistent insight — for structural, thermal, and fluid problems.

  • Faster structural and thermal optimisation across design variants
  • Early detection of failure zones, hotspots, and flow issues before physical prototyping
  • Rapid iteration of geometry, loads, and boundary conditions at simulation speed
  • More confident design decisions across cost, performance, and reliability
Vision
"To become the AI-native simulation layer powering industrial digital twins — starting with internal airflow and thermal systems, and expanding into broader structural and multiphysics domains."
phinen is not just a faster solver. It is a foundational intelligence layer that transforms how physical systems are understood, iterated, and optimised across industries.
Phase 1 · Now
Structural FEA · Thermal & CFD · Multiphysics simulation
Phase 2 · Next
Comparative analysis · Multi-variant optimisation workflows
Platform
Cloud-native · Scalable collaboration · API-first
End State
Live digital twin — persistent simulation layer for physical assets
Why Now
✓ AI-based surrogate modelling has matured enough for engineering-grade accuracy
✓ GPU infrastructure is widely accessible — inference is now viable in SaaS
✓ Digital twin adoption is accelerating across all target industries
✓ Legacy simulation stacks were not designed for foundation models or LLM-guided workflows

Get Started

Simulate Faster.
Design with Confidence.

Join the early access programme and get phinen running on your geometry — structural, thermal, or fluid — within 24 hours.

No spam · No commitment · We'll reach out within 24 hours

Structural FEA Thermal & CFD Multiphysics Automotive & EV Aerospace & Defense Medical Devices Healthcare & MedTech Robotics Supply Chain Energy & Power