Build the research engineering platform a Type I civilization relies on.
Three live panels share one shader pipeline and one state timeline. The system is geometry-aware end-to-end: metrics and operators are first-class, every frame is replayable, and every derived field is reproducible from logged seeds and inputs. Target: 60 fps on a laptop GPU, deterministic replay within 1e−6 relative error for derived scalars, and bounded numerical drift under camera and timestep changes.
What you will build
If you join MMI, you will build agent (intelligence) systems that operate over plasma-class dynamical systems and enterprise-scale processes. This work replaces toy abstractions with geometry, operators, and control-relevant structure. Example build areas:
- Intelligences for plasma-class systems Build intelligences that ingest fusion, electromagnetic, or climate fields, infer latent structure, and propose interventions. Implement GR- and EM-derived operators, validate them against reference solvers, and expose them through a world-model-backed interface suitable for real-time interaction.
- Visualizations as intelligent-addressable manifolds Treat every visualization as a first-class observation manifold. Define metrics, charts, and bounds so intelligences can trace geodesics, estimate curvature, detect regime shifts, and request new slices programmatically rather than relying on static dashboards.
- Closing the plasma–organization loop Apply stability and control principles from plasma physics to human organizations. Build intelligences that detect attention sinks, coordination turbulence, and information shear, and test interventions using the same replayable, bounded-action framework used for physical systems.
- Automated scientific reasoning Design intelligences that propose experiments, tune PDE parameters, discover symmetries, and evaluate reduced models against real data. Emphasis is on traceability, numerical stability, and falsifiable gains rather than paper-only results.
- Expanding enterprise phase space Model enterprises as dynamical systems with controllable attractors. Build tools and intelligences that map reachable phase space, identify unstable trajectories, and recommend actions that increase controllable variety under real operational constraints.
You will not receive a fixed spec. You will define the ontology, interfaces, and safety constraints, ship incrementally, and measure impact through reproducibility, stability, and increased control over complex systems.
Who this is for
Likely a fit if:
- ✓ You have read actual QED/QFT or GR texts and can distinguish canonical results from speculation.
- ✓ You can switch coordinate systems mentally and know what transforms and what does not. You are comfortable with metrics, embeddings, flows, and practical approximations.
- ✓ You have built nontrivial visual systems (custom shaders, scientific visualization kernels, simulation dashboards, or similar).
- ✓ You can go from a concept sketch to a running prototype quickly and treat wrong-but-illuminating artifacts as part of the process.
- ✓ You value modular abstractions and reproducibility more than polishing one perfect demo.
- ✓ You are motivated by global engineering and using models to compress search, test hypotheses, and extend your own bandwidth.
- ✓ You treat intelligence and modeling as finite-precision dynamical systems and can reason about stability, safety, and capacity in that frame.
Not a fit if:
- ✕ You cannot tolerate incomplete or slightly incorrect visualizations as scaffolds.
- ✕ You prefer fixed, linear ticket execution instead of open-ended problem solving.
- ✕ You are uninterested in physics, field theories, simulation fidelity, or control theoretic reasoning.
- ✕ You avoid debugging across layers such as geometry, numerics, rendering, optimization, and ML models.
We are hiring Intelligence Systems Engineers.
Over the next cycles we expect to staff the equivalent of 3 to 12 full-time agent engineers with overlapping skill sets:
Titles are flexible. The core requirement is that you and your intelligences can span at least three of these archetypes and operate as a unified human+intelligences engineering loop.
Location and work mode
- Base: Customer-site dependent
- Remote: Considered for exceptional candidates with strong prior signal and the ability to work with minimal supervision and high bandwidth async communication
Short stints in person are encouraged even for remote hires. The work is faster when we can stand in front of the same screen and argue about field lines.
How to apply
What world model will you build to understand and extend this page. Show us how you think, how you reason about geometry, and how you work with automated intelligences inside a controlled development loop.
- Submit your resume and links.
GitHub, personal site, or any prior visual, simulation, or physics work is useful.
- Automated screen.
You will receive a short coding challenge centered on extending or improving the shader based Earth–Moon–electron demo. The focus is clarity and iteration. You will define a small interface, make a focused change, and include a quick self-check or invariant that keeps the result stable.
- Agentic workflow check.
The challenge includes one step where you specify how a Planner, Implementer, and Reviewer intelligences would operate on your change. We want to understand how you design boundaries, supervise automated contributors, and verify correctness.
- Deep technical conversation.
If the challenge goes well, we will explore your physics intuitions, your engineering habits, your approach to world models, and how you coordinate with intelligences in a high-velocity research loop.
After clicking, send a LinkedIn message with a short intro and 2–3 links (GitHub, portfolio, or 1–2 builds).
This link routes to LinkedIn.