Founding Engineer

Silmaril

Security for agents that self-improves.

We are backed by Y Combinator and based in San Francisco.

The Opportunity

Silmaril brings research and production engineering into the same loop.

You will choose important technical problems, develop hypotheses, build evaluations, run experiments, and carry successful ideas into production. Your work could reduce inference latency, develop faster and more capable security models, create autonomous red-team systems, or turn production evidence into new training methods.

You will operate with the scope of technical staff, moving across disciplines and owning foundational work from first principles through production.

AI Native Engineering

AI has changed what a single great engineer can build. We are looking for someone who has internalized that shift more completely than almost anyone.

You have built a code factory around yourself: parallel agents, durable context, automated evaluation, review, testing, and observability. You can direct several ambitious workstreams while retaining command of the architecture and underlying details.

Your output curve changed with advances in AI. You can build hundreds of thousands of lines of coherent, reliable software and execute at a scale previously associated with entire teams.

At Silmaril, you will apply that leverage to one of the defining technical problems of the AI era.

You Will Own

Research direction

Identify the technical problems with the greatest potential to advance the product. Design rigorous experiments, build the necessary infrastructure, and use evidence to determine what advances.

Models and inference

Improve model capability, accuracy, latency, efficiency, and deployability under demanding production constraints.

Adversarial research

Build autonomous red-team agents, attack-generation systems, and evaluation harnesses that continuously expose new failure modes.

Learning systems

Turn attacks and production outcomes into better datasets, evaluations, models, and defenses.

Core architecture

Design the infrastructure and decision systems at the center of Silmaril.

Production systems

Own correctness, reliability, observability, isolation, deployment safety, and recovery in consequential environments.

Technical standards

Help define how Silmaril conducts research, evaluates progress, builds software, and operates critical systems.

You

• You have five or more years of professional software engineering experience.
• You have built and operated consequential production systems at a technically ambitious company.
• Your work demonstrates unusual scope, speed, and technical depth. You can move from a research question to an experiment, architecture, implementation, deployment, and operation.
• You know how to structure agent work, provide the right context, evaluate results, resolve uncertainty, and concentrate your attention on the highest-leverage decisions.
• You have made something exceptional exist from an ambiguous beginning and taken responsibility for how it performs in the real world.
• You want broad ownership, close collaboration, and technical problems whose solutions could define an industry.
• You want to build in person with a small, exceptional team in San Francisco.

Bonus

• Experience with model training and evaluation, inference optimization, adversarial research, distributed systems, production ML, security engineering, or developer infrastructure.
• A record of technical research, open-source work, or ambitious independent projects.
• An exceptional GitHub profile or body of AI-native work that shows a clear increase in scope and output as models advanced.