Founding AI Engineer

Founding AI Engineer (AI + Production)

New York City (5 days on-site) · Top of market + equity + benefits

TL;DR: Build AI that accelerates nuclear deployment. Own AI production from evals to fine-tuning. Push the frontier on physics models, world models, and AI-accelerated simulations. High-leverage IC role with founding-level impact.

The Mission

Everstar builds the intelligence layer that makes nuclear power actually deployable —collapsing regulatory and manufacturing timelines from years to months. Gordian already powers engineering and compliance work for utilities, advanced reactor companies, and hyperscalers. We pair deep nuclear domain expertise with frontier AI and move with startup speed.

Now we need a Founding AI Engineer to turn research breakthroughs into production systems that ship—and push beyond LLMs into physics-informed AI, world models, and simulation acceleration.

You’ll be joining the Apollo Team of Nuclear . You’ll build alongside engineers from Tesla, SpaceX, Lockheed Martin, Google, and Microsoft. You’ll learn from nuclear and national security experts who cut their teeth at the Nuclear Regulatory Commission, CIA, and NuScale.

The Role (reporting to the CEO)

Not a researcher. Not a prompt engineer. This is a production-first role.

You'll own the AI stack end-to-end —from eval frameworks to fine-tuning pipelines to agent orchestration. But you'll also push the boundaries of what AI can do for nuclear: AI-accelerated weather simulations, design safety analyses powered by physics models, and world model applications that transform nuclear operations.

Think 70% building production systems / 30% frontier R&D with access to Microsoft and NVIDIA's latest tools through our first-party partnerships and a large AI research budget to experiment aggressively.

Most weeks you'll be shipping new model capabilities, debugging eval failures, and scaling inference—then immediately applying what you learned to the next sprint. Some weeks you'll be prototyping physics-informed models, running GPU-accelerated simulations, or collaborating directly with NVIDIA and Microsoft researchers.

You will:

Build production AI agents: power Gordian Search, Research, and Compose with outputs that are truthful, complete, and auditable—because in nuclear, "mostly right" isn't good enough.

Design eval infrastructure: create benchmarking suites that catch regressions before customers do; instrument quality metrics that actually matter.

Own fine-tuning pipelines: generate synthetic data, run ablations, and ship domain-adapted models that outperform off-the-shelf LLMs on nuclear regulatory tasks.

Push the frontier (R&D):

AI-accelerated weather simulations for site qualification and environmental impact assessments—replacing months of modeling with hours

Physics-informed design safety analyses using world models that reason about thermal hydraulics, neutronics, and structural integrity

Vision + physics models for automated document analysis, construction monitoring, and operational anomaly detection

Agentic workflows that compound over time, learning from each regulatory submission to improve the next

Leverage NVIDIA partnership: work directly with NVIDIA's research team to access cutting-edge tools (NeMo, Modulus, Omniverse) and contribute to the future of AI for critical infrastructure

Set technical direction: you're early enough to shape how we think about model selection, prompt design, guardrails, physics-AI integration, and the entire ML ops stack

Mentor and lead: as the team scales, you'll hire and guide other AI engineers—but first, you'll prove the playbook yourself

A sample week: debug why Research citations dropped 8%; ship new fine-tuned model for compliance drafting; design eval suite for multi-document reasoning; prototype physics-informed model for thermal analysis; pair with fullstack engineer to optimize inference latency; attend NVIDIA collaboration session on world models; read three ML papers and implement one idea.

What You've Done

3–8 years building production ML/LLM systems—RAG, fine-tuning, evals, agent orchestration. You've shipped models that users depend on daily.

Mastery of the stack: Hugging Face, LangChain, vector databases, prompt engineering, and modern LLM ops. You know when to use off-the-shelf and when to build custom.

Rigor with evals: you've designed benchmark suites, tracked model quality over time, and know how to measure what matters (not just what's easy).

Leadership DNA: you've owned outcomes, not just tasks. You've set technical direction, mentored teammates, or led cross-functional projects.

Bonus points:

Experience with physics-informed neural networks , scientific computing, or simulation acceleration

Published research in ML/AI, contributions to open-source ML frameworks

Deep familiarity with NVIDIA tools (NeMo, Modulus, CUDA optimization)

You're the person who reads Arxiv papers on weekends and immediately wants to implement them

Background in physics, engineering, or computational science

No nuclear background required—only the hunger to build AI that matters and push the boundaries of what AI can do for physical systems.

Who You're Building For

This isn't benchmarks for benchmarks' sake. Your models will directly help:

Nuclear operators keeping 20% of U.S. electricity safe and reliable

Advanced reactor developers navigating regulatory approval for next-gen designs—and using AI-accelerated simulations to optimize designs in days, not months

Licensing teams drafting safety analyses that take months today, hours tomorrow—powered by physics models that understand first principles

Site qualification teams running environmental and weather analyses that currently require expensive consultants and 6+ month timelines

And the second-order effects matter even more:

Nuclear unlocks the energy needed for AGI/ASI —advanced AI requires unprecedented power.

AI accelerates nuclear deployment —breaking the regulatory bottleneck that's held back clean energy for decades.

The tokens you generate translate into safer infrastructure and a livable planet.

What’s at Stake

🔥 If we succeed: We unlock nuclear at scale, power the AI revolution with clean energy, and collapse licensing timelines from years to months. The models you build help humanity leap toward AGI on a sustainable foundation. Your physics-informed AI becomes the standard for how critical infrastructure is designed and operated.

❄️ If we fail: Nuclear stays bottlenecked in decades-old processes, AI's energy demand outpaces clean supply, and we miss the window to align technological progress with climate survival. The frontier AI capabilities remain academic curiosities instead of deployment accelerators.

What Success Looks Like (90 days)

Shipped ≥3 major model improvements to production (better evals, new fine-tuned model, or agent capability).

Eval framework is instrumented and running continuously ; you catch quality regressions before customers report them.

Inference latency reduced ≥30% or accuracy improved ≥15% on key benchmarks.

Prototype ≥1 frontier capability (physics model for safety analysis, weather simulation acceleration, or world model application) that shows clear customer value.

You've set the technical roadmap for AI engineering and the team trusts your judgment.

At least one system you built (eval suite, fine-tuning pipeline, or agent orchestration) is now core infrastructure the company depends on.

Resources at Your Disposal

NVIDIA & Microsoft first-party partnership: Direct access to Microsoft & NVIDIA research team, early access to new tools (NeMo, Modulus, Omniverse), and collaboration on frontier AI applications

Large AI research budget: Aggressive compute allocation for training runs, experiments, and frontier R&D—no need to beg for GPU credits

Latest NVIDIA hardware: Access to H100s, GH200s, and future architecture