Job details
About this role
Role overview Contribute to structured generation technology that constrains language model outputs by deciding which tokens are allowed at each decoding step. The team compiles JSON Schemas, regular expressions, grammars, function definitions, or policies into a fast decision engine that runs inside the inference loop, enabling reliable and safe agents. This role focuses on the parsers and compilers at the core of the system, blending research and production engineering.
Responsibilities - Design and develop high-performance parsers and data structures that integrate with large language models - Research advanced parser and compiler technology and symbolic methods - Ensure reliability and efficiency of methods through rigorous testing and optimization - Stay current with the latest relevant advancements in language model technology
Requirements - Professional-level proficiency in Python plus practical experience with a lower-level language such as C, C++, or Rust - Hands-on experience with parsers such as Earley or LALR(1), formal grammars, finite-state machines, and efficient data structures, plus the ability to do research and development in those areas - Familiarity with the relevant public technical report on the technology and its open-source implementation - Demonstrated commitment to simplicity, testing, and automation - Clear, concise, and consistent written communication
Benefits and work setup - Fully remote with most of the team in the US and EU; requires at least three hours of daily overlap with 8am–1pm Eastern Time - Market-rate salary plus stock options, health and dental insurance, and a 401(k) for US employees - GPU hardware provided if needed - Written-first remote culture with occasional organic discussions