Software Engineer, Full-Stack ML Infrastructure

Description

The company is a frontier AI lab based in Palo Alto, founded in 2025, that builds a verification layer for AI. It converts complex human knowledge such as tax codes, legal rules, clinical guidelines and security protocols into a formal representation, so each AI answer can be traced, questioned and proven.

Its architecture combines foundation models trained to formalize and reason with a symbolic world model written in Lean 4. The founding team came from Google, DeepMind and Glean and brings together AI researchers and formal methods experts.

Responsibilities

• Build the tools and systems that help engineers and researchers develop, test, deploy and run their work more effectively.
• Start with DevOps and MLOps: improve development environments and deployment workflows, make experiments and evaluations reproducible, and help teams see how AI systems behave, perform and cost. The work spans interfaces, backend services and infrastructure automation, not only scripts or dashboards.
• Early on, the scope covers the whole company. Over time the work may extend to shared capabilities such as identity and access, LLM gateways, API-key management, training and inference platform upkeep, usage metering and internal applications. You will work directly with the people using these systems and replace recurring friction with reliable self-serve tools.
• Treat internal tools as products: useful, easy to understand, well maintained and designed around the workflows of the people who rely on them.
• Improve how software is built and shipped. Create repeatable local environments, automated tests, CI/CD pipelines and deployment workflows. Make configuration, secrets, releases and rollbacks easier to manage safely.
• Build tooling for the AI development lifecycle. Create workflows for running experiments and evaluations, versioning datasets and configurations, tracking results and artifacts, and deploying or integrating model services. Make runs easier to reproduce, compare and investigate, and help researchers move from a recipe to a running job and from a failed job to its cause.
• Build full-stack, self-serve applications with interfaces and APIs that let teams launch jobs, manage configurations, inspect results and troubleshoot failures without repeated manual help.
• Connect logs, metrics, traces and usage data so teams can diagnose issues, find bottlenecks and see where time and compute go.
• Help build and integrate identity, access controls, LLM routing, API keys, budgets and usage metering across the company's products and internal workflows.
• Evaluate open-source and managed solutions, and extend them where they fit. Build custom software only where it fills a real gap instead of rebuilding existing infrastructure.
• Work with engineers and researchers to find recurring problems, set priorities and measure whether tools cut setup time, manual work and failed runs. Maintain tests, documentation and operational guidance for what you ship.

Requirements

• At least 2 to 3 years of software engineering experience, including hands-on work with backend services, full-stack applications, internal tools or deployment automation.
• Solid skills across the stack. You can build services and automation in Python or TypeScript, work with APIs and databases, and build usable interfaces with React or a similar framework.
• Practical cloud and deployment experience. You have deployed applications with containers on a cloud platform such as AWS, worked with CI/CD, and used logs or metrics to investigate problems.
• Some hands-on exposure to AI or ML workflows, such as integrating model APIs, running evaluations, tracking experiments or operating model-backed applications, gained through work or substantial projects. You do not need to be an ML researcher.
• A product mindset for internal users. You care about clear interfaces, sensible defaults and useful documentation, and you work to reduce the steps someone needs to get their work done.
• Sound engineering judgment and a collaborative style. You write maintainable code, test what matters, ask for review and explain trade-offs. You can own a scoped problem and know when to ask for help.
• Especially strong candidates may also have the following. These are not required:
• Experience with experiment tracking, evaluation pipelines, workflow orchestration or model observability using tools such as MLflow, Langfuse or Temporal.
• Experience with infrastructure as code, container orchestration or telemetry tooling, such as Terraform, Kubernetes or OpenTelemetry.
• Experience running GPU workloads on Kubernetes, including job queueing and preemption, multi-node training with NCCL over EFA or InfiniBand, GPU node health and driver management, or serving models with vLLM or SGLang.
• Experience with authentication, role-based access, API-key management, image provenance, LLM gateways, usage and cost tracking, or hardening public endpoints.
• Experience adapting open-source software or building developer tools that other engineers or researchers use regularly.
• Not having experience with every system above is fine. Strong software fundamentals, practical problem-solving and interest in DevOps and MLOps matter more than matching every tool.

Conditions

This is an early-stage role with scope across the company.