Artificial Intelligence Engineer

Neuron Industries (YC S26); Gardena (CA); On-site; USD 180,000 - 280,000; Full time

Benefits offered by this job

100% employer-paid Medical, Dental, &

Vision insurance

Free daily catered lunch & dinner

Job summary

Neuron Industries is hiring for a founding AI platform lead in Gardena, CA. You will own the AI fabric, agent fleet, eval stacks, and open-weight model strategies to deliver production-grade AI for industrial automation.

You will design how models plan, call tools, and recover from failures, while steering cost, latency, and performance across the company’s AI stack.

Qualifications

• Shipped LLM-powered systems used by real users.
• Built evaluation suites that catch regressions before release.
• Implemented context management like retrieval, memory, pruning.

Responsibilities

• Own our agent building end to end: workflows in our IDE and internal agent fleet.
• Own the eval stack and CI gates with calibration checks.
• Develop harnesses for tools, sandboxes, and recovery paths used across the company.
• Design context strategies: retrieval, memory, and token budgets.
• Train and deploy open-weight models (SFT, LoRA, RL-style tuning) and serve them in production.

Skills

LLM systems

Eval suites

Context engineering

Open-weight models

Backend engineering

Tools

vLLM

SGLang

Quantization

MCP-style interfaces

Job description

Own AI at Neuron: product agents, the internal agent fleet, the eval stack, harness and context engineering, and open-weight training and serving. A founding role, not a first AI job.

But we cannot manufacture our abundant future while the reactors, fabs, and production lines that make everything we use still run on architecture locked in the 90s.

Neuron Industries is the industrial technology company for the AI era. We are rebuilding the entire industrial automation tech stack, from hardware to runtime to the AI fabric, without sacrificing a minute of uptime. Controls projects that took weeks now take hours.

Come help us build the golden age of American manufacturing.

The role

The AI fabric is yours: the agents in our products, the agent fleet that runs the company, the evals that keep both honest, and the models underneath. We need someone who has shipped this for real and can prove it.

What you’ll do

• Own our agent building end to end: the AI workflows in our IDE that turn weeks of controls work into hours, and the internal agent fleet that runs research, sales ops, and the back office.
• Own the eval stack: task suites, regression gates in CI, LLM-as-judge with calibration checks, and the online metrics that catch a degraded agent before a customer does. If it isn't evaled, it doesn't ship.
• Do harness engineering: the loops, tools, sandboxes, and recovery paths agents run inside. You design how they plan, call tools (MCP-style interfaces), fail, and retry, and you make those harnesses reusable across the company.
• Do context engineering: retrieval, memory, compaction, caching, token budgets. Getting the right information in front of the model at the right time is most of the game, and it's yours.
• Train open-weight models where they beat the frontier for us — SFT, LoRA, preference and RL-style post-training for controls tasks — and serve them (vLLM/SGLang-class, quantization) for customers whose data can never leave the building.
• Own the model landscape and the AI spend: frontier vs open-weight, cost, latency, capability, re-picked when the answer changes. The whole company works AI-first, and you're the one who makes the rest of us sharper at it.

What you’ll bring

• You've shipped LLM-powered systems that real users depended on, and you can walk us through one end to end: the harness, the context strategy, the eval suite, what broke, and what you changed.
• Evals are reflex. You've built suites that caught real regressions before users did, you know why LLM-as-judge drifts and how to calibrate it, and you don't trust a demo.
• Real context engineering: retrieval that holds up in production, memory and compaction schemes, and you know why a bigger context window doesn't save you.
• You've post-trained open-weight models (SFT, LoRA/QLoRA, DPO or RL-style preference tuning) and served them in production. And you know when fine-tuning is the wrong answer.
• Strong software engineering underneath. Agents are distributed systems: async, retries, state, observability. You reason about cost and latency like a backend engineer, not a notebook user.
• The field moves monthly and you like that. You read the model releases, re-test your assumptions, and change your mind on evidence.

Nice to have

• You've built RL environments or verifiable-reward training loops for agents.
• GPU infrastructure: cluster scheduling, distributed training, inference optimization.
• You've made models good at a niche domain (code, legal, hardware) where the frontier was weak.
• Open-source work in the agent, eval, or serving ecosystem that people actually use.

Compensation

$180,000–$280,000 base, plus meaningful founding-level equity. The range is wide on purpose: this is a founding seat, and where you land depends on what you have actually shipped and can prove.

• 100% employer-paid Medical, Dental, and Vision insurance (premium PPO and HMO options)
• Paid vacation, sick leave, and company holidays
• Free daily catered lunch and dinner

Neuron Industries is an equal opportunity employer: we hire on ability, hunger, and what you can build. And if this role excites you but your experience doesn't line up perfectly with every bullet above, apply anyway. The bullets describe the work, not the prerequisite for success.

Location:

Gardena, CA — on-site, five days a week.