Research Engineer

Stealth AI Research Lab; Cambridge (MA); On-site; USD 120,000 - 190,000; Full time

Job summary

Stealth AI Research Lab is seeking an AI Research Engineer to advance Cortex, exploring learning with brain-inspired methods and continuous improvement. You will work with scientists, engineers, and collaborators to push the boundaries of sample-efficient AI and explicit knowledge representation.

You will publish findings, develop benchmarks, and contribute to product alignment. The role emphasizes long-horizon research, novel architectures, and meaningful scientific impact in a seed-stage

Qualifications

• Demonstrated research track record with first-author publications or equivalent open-source impact.
• Public record of original ideas beyond applying known methods.
• Strong mathematical foundations in probability, optimization, and linear algebra.
• Hands-on experience training/evaluating neural networks, including LLMs, with PyTorch or JAX.
• Solid experimental discipline with baselines, ablations, and transparent reporting.
• Experience building software with LLMs or agentic systems used by others.
• Ability to propose and test sharp hypotheses and drop weak ideas quickly.
• Excellent written and verbal communication; proactive collaboration.

Responsibilities

• Develop Cortex iterations and benchmark against long-horizon challenges in agentic reasoning and continual learning.
• Publish benchmarks and evaluations exposing learning failure modes in AI systems.
• Research brain-inspired learning and knowledge representation; integrate neuroscience findings into Cortex.
• Improve agentic capabilities: planning, search, tool use, query understanding, safe abstention.
• Apply post-training techniques to improve sample-efficient continual learning.
• Author papers and technical reports; influence research direction.
• Transition experiments into production with clear tests and regression gates.

Skills

ML research

Neuroscience

Mathematics

Software with LLMs

PyTorch/JAX

Experimentation discipline

Scientific writing

Communication

Tools

PyTorch

JAX

Job description

We are a team of neuroscientists, mathematicians, AI researchers, and engineers on a mission to learn from the human brain and to build an AI that can learn continuously, in a sample-efficient way, and reason about its knowledge. We are a seed-stage team with early commercial traction building Cortex . Cortex is in use by customers today, and we're only just beginning in terms of research and capabilities.

About Us

We are a team of neuroscientists, mathematicians, AI researchers, and engineers on a mission to learn from the human brain and to build an AI that can learn continuously, in a sample-efficient way, and reason about its knowledge. We are a seed-stage team with early commercial traction building Cortex . Cortex is in use by customers today, and we're only just beginning in terms of research and capabilities.

About the role

We're looking for an AI research engineer to bring ideas and execution toward understanding natural intelligence and implementing it in silico. You'll develop the next iteration of Cortex and test it against hard challenges (like ARC-AGI and others). You'll also help design and publish new algorithms, approaches, and capabilities for Cortex, inspired by the brain and specifically focused on sample-efficient continuous learning and explicit knowledge representation. You'll create and publish new benchmarks that set the bar for the capabilities of Cortex and all next-generation thinking machines.

Your near-term research focus will be on post-training techniques, novel reward models, differentiable approaches to explicit memory stores, harness improvements, native multimodality, and direct application of existing neuroscience research to Cortex.

Longer-horizon research goals we expect you to participate in include exploring non-transformer architectures, novel information encoding approaches, and beyond. We expect to push the envelope from "interdisciplinary" work to "a new, named field of science" and you will be part of it.

What you'll do

As a Research Engineer, you'll bring both engineering discipline and research taste. You will work with our research scientists and mathematicians to explore and publish. You'll work with our engineers to translate this into product. Your primary outputs will be both publications and code artifacts, about a 50/50 split. We do not have an open source strategy yet; come help define it.

• Develop the next iteration of Cortex, and benchmark it against hard, long-horizon challenges in agentic reasoning, continual learning, and knowledge use.
• Participate in the design and publishing of new benchmarks and evaluations that expose failure modes in how AI systems learn, represent, and use knowledge.
• Research brain-inspired approaches to learning and knowledge representation: how knowledge is structured, linked, updated, resolved, and forgotten, drawing on neuroscience and cognitive science, as well as modern AI research.
• Improve the agentic harness around Cortex: planning, search, tool use, query understanding, and grounded abstention.
• Apply and evaluate post-training techniques (fine-tuning, reinforcement learning, distillation) where they improve sample-efficient, continual learning.
• Author papers and technical reports, and help shape our research direction.
• Land research in the product, working with the engineering team to take ideas from experiment to production with tests and regression gates.

You may be a good fit if you are strong in some of the following

• Have a research track record in machine learning, neuroscience, physics, mathematics, computer science, or another quantitative field, with ideally first-author publications at top venues such as NeurIPS, ICML, ICLR, or CoLLAs.
• Have a public record of original ideas, shown by papers, open-source work, or results that others have built on, not only the application of known methods.
• Have strong mathematical foundations: for example, in optimization, probability, dynamical systems, information theory, and linear algebra.
• Have deep hands-on experience training and evaluating neural networks, including LLMs and post-training methods, for example using PyTorch or JAX.
• Have strong experimental discipline: careful baselines, ablations, and honest reporting of negative results.
• Have built software using large language models, including agentic systems, ideally software that was really used by others.
• Have research taste: you pick problems worth solving, form sharp hypotheses, and drop weak ideas quickly.
• Communicate beautifully in writing and proactively in conversation. Most problems are not technical and we need you to help solve those ones too.
• Are curious about how biological learning works, and want to turn those ideas into working systems.

• Published research on continual or lifelong learning, sample-efficient learning, memory architectures, or brain-inspired machine learning.
• Experience with non-transformer architectures such as state-space and recurrent models, energy-based or Hopfield-style models, spiking networks, or attractor dynamics, as well as SAEs, dictionary learning, or more old-school ML methods.
• A computational neuroscience background in areas like hippocampal function, complementary learning systems, consolidation, or predictive coding.
• Experience owning full software projects that hold up in high-stakes production environments: auth and RBAC, security and privacy, latency-sensitive paths, to name a few, and the reliability habits (monitoring, safe rollouts, on-call) that come with real customers.
• Real-life experience with reinforcement learning, reward modeling, and large-scale post-training.
• Benchmarks or evaluations you've designed that other researchers have adopted.
• Open-source projects or code that has gained traction in the research community, beyond a few stars.

We're looking for exceptional people, and we know few candidates will match every line above. If this is the work you want to do and some of it describes you, we are excited about the prospect. Pedigree and experience checkboxes often push away underrepresented groups. That's not what we're trying to do. We're signaling that we expect the best and we care about the quality of your thinking. This space is exciting because only the quality of our ideas and execution bottlenecks us.

equal opportunity employer.