HomeDocsAPICopilotBlogVerificationPricingCareersApply now
We're hiring · founding engineers

Build the verified simulation engine
for engineering AI.

physicsbase is a finite-element engine. Engineering agents call it to author, mesh, simulate, and check real physics. It checks every result against an exact solution or a benchmark. We now make our first two hires. Both are founding engineers. Both are deeply technical. Both own part of the core. Read the two roles below, then apply.

Roles open
2 founding engineers
Equity each
3%
Location
Remote (NA overlap)
Type
Full-time
01 · Computational Mechanics Engineer 02 · Simulation Platform Engineer Apply now →

To apply, do these steps. Read a role. Click its Apply for this role button. Then add your details in the form and click Submit application. You can also open the form now.

Role 01 · Solver & Verification

Founding Computational Mechanics Engineer

Engineering Remote · NA overlap Full-time Equity 3%

You own the numerical core. The core has the finite-element formulations, the material models, and the solvers. You also own the evidence that proves each result is correct. physicsbase already has nonlinear mechanics, composites, contact, explicit dynamics, CFD, acoustics, and coupled multiphysics. A closed-form check or a benchmark controls each capability. You make this library larger. You keep the verification standard high as it grows.

What you do

  • You design and build finite elements: solids, shells, and structural members. You add large-deformation kinematics, contact, and constraints. You show the correct convergence rate for each one.
  • You build material models: plasticity, viscoelasticity, hyperelasticity, damage, and rate effects. You give each model a consistent tangent. You show quadratic Newton convergence.
  • You write the checks that prove each capability before it ships. You use closed-form solutions, manufactured solutions, and published benchmarks. You keep every check green in CI.
  • You improve error estimation, adaptivity, and time-step stability. The engine then tells an agent when a mesh or a step size is too coarse.
  • You profile and speed up the hot paths: assembly, sparse factorization, and element loops. Real models then solve in a usable time.

Deep technical requirements

  • You know the finite element method well: weak forms, isoparametric mapping, quadrature, locking remedies, and the sources of discretization error.
  • You know nonlinear solid mechanics: finite-strain kinematics, the deformation gradient, objective stress rates, return-mapping algorithms, and consistent tangents.
  • You know numerical linear algebra for FE: sparse direct and iterative solvers, conditioning, modal and buckling eigenproblems, and the stability limits of explicit and implicit integration.
  • You verify by instinct. You use manufactured solutions, mesh-refinement studies, and energy and momentum balances. You can cite Timoshenko, Roark, NAFEMS, or MacNeal & Harder for a check.
  • You write production scientific Python (NumPy and SciPy). You read and write vectorized element kernels. C, C++, Rust, or GPU work is a plus, not a requirement.

Experience

You have a PhD or an MSc in computational mechanics, an engineering field, or applied mathematics. You have also shipped real code. An equal record from industry or open source also counts (about 4 or more years of hands-on FE work). We value a solver you can defend line by line more than a specific degree.

Apply for this role Founding equity: 3% · Remote · Full-time
Role 02 · Agent-Native Infrastructure

Founding Simulation Platform Engineer

Engineering Remote · NA overlap Full-time Equity 3%

You turn a verified solver into a platform. Agents and engineers then use it at scale. This work covers the API, the MCP tool layer, the compute and job system, the CAD and mesh import pipeline, and the data plane for accounts, usage, and results. physicsbase already runs over REST, MCP, and a Python SDK. You make it fast, reliable, and observable. You make it ready for large geometry and long solves.

What you do

  • You own the REST, MCP, and SDK surfaces. You keep the schema, the versioning, and the error contracts clean. An agent then makes a correct call on the first try.
  • You build the compute plane: an async job system, a worker pool that scales past one process, backpressure, timeouts, and result storage for large and long solves.
  • You harden the CAD-to-mesh-to-simulate pipeline. You handle STEP, IGES, and mesh import, meshing controls, and node budgets. Files of real size then flow through cleanly.
  • You design the data layer for accounts, API keys, usage metering, and results. You keep a strict boundary between product analytics and user data.
  • You own reliability: CI/CD, observability, rate limiting, safe deploys, and the performance work that keeps p95 latency honest.

Deep technical requirements

  • You write strong Python backends (FastAPI or similar). You have real API judgment: idempotency, pagination, versioning, and machine-readable error contracts.
  • You know distributed systems: queues, worker pools, concurrency, caching, backpressure, and graceful degradation under load.
  • You do data engineering: SQL schema design, migrations, transactional integrity, and a clean split between PII and telemetry.
  • You run production systems: containerized deploys, CI/CD, metrics, tracing, logging, and load tests you can point to.
  • Bonus: you have used agent tools (MCP or function-calling), CAD or meshing kernels (gmsh or OpenCASCADE), or scientific compute. This helps you talk with the solver team.

Experience

You have about 4 or more years on production backends or developer platforms. You have taken a latency-sensitive or compute-heavy system from prototype to a reliable service. Shipped and operated systems matter more than a degree. Work at the seam of infrastructure and numerical or ML workloads is a strong signal.

Apply for this role Founding equity: 3% · Remote · Full-time

Apply

Use this form for both roles. Pick the role. Tell us what you built. Show us the evidence. We read every application ourselves. You get a real reply from a person.

Give a public link: a PDF, a personal site, or a shared drive. This form does not accept file uploads.
Use 80 to 5000 characters. Be specific. Show the evidence, not only the result.
Use 40 to 2500 characters.