Software Engineer, Tokens and Prompt Structures

Details

Anthropic's Encodings Infra team maintains the libraries used to encode text and multimodal data into a form that Claude can consume. The team also defines Claude's prompt structures, including how user turns are represented, how Claude calls tools, and how it receives tool results.

The Software Engineer will own the design and maintenance of these libraries, focusing on intuitive APIs, performance, and robust abstractions. The role spans systems across the codebase, from pretraining and fine-tuning to the API, and involves close collaboration with researchers and engineers to move new encoding ideas from experimentation to production.

Responsibilities

• Maintain and improve encoding libraries used by Anthropic engineers and researchers.
• Run experiments to determine effective ways to provide structured data to Claude without causing confusion.
• Design data structures and abstractions that hide encoding details from most of the organization while supporting advanced users.
• Adapt encoding libraries to support emerging research directions and enable production deployment.
• Optimize encoding performance across dependent systems.

Requirements

• 5+ years of software engineering experience, including meaningful experience maintaining libraries, SDKs, or developer-facing APIs.
• Familiarity with machine learning terminology and large language model architecture.
• Experience conducting complex refactors in large codebases.
• Strong communication skills and an ability to collaborate closely with researchers and engineers.
• Results-oriented approach, flexibility, and a focus on impact.
• Willingness to take on work beyond the formal job description.
• Interest in the societal impacts of the work.
• Minimum education of a bachelor's degree or equivalent education, training, or experience.
• Relevant field of study demonstrated through coursework, training, or professional experience.

Preferred Experience

• Tokenizers or other text and data encoding systems.
• Maintaining a widely used library over an extended period.
• Performance optimization.
• Python and/or Rust.
• Reinforcement learning or model training infrastructure.

Representative Projects

• Working with a research team to ship a new multimodal data type, such as audio or video, to production.
• Redesigning a core abstraction to enable changes to Claude's data encoding without breaking downstream teams.

Compensation

• Annual salary: $320,000–$405,000 USD.

Logistics and Benefits

• Hybrid policy: Staff are expected to work from one of Anthropic's offices at least 25% of the time; some roles may require more office time.
• Anthropic sponsors visas, although sponsorship cannot be guaranteed for every role or candidate. The company retains an immigration lawyer to assist with visa processes.
• Benefits include competitive compensation, optional equity donation matching, generous vacation and parental leave, flexible working hours, and office collaboration space.