Founding Technical Staff - AI Safety Research

Sampura Research; Greater London; Presencial; GBP 100.000 - 290.000; Jornada completa

Ventajas ofrecidas por este puesto de trabajo

Visa sponsorship

Descripción de la vacante

Sampura Research, an AI safety research non-profit based in London, seeks founding Members of Technical Staff to shape the direction of research and engineering and own end-to-end problems within our agenda.

You will lead evaluation design, implement methods to improve judge performance, and build scalable infrastructure for research, publications, and benchmark releases while collaborating with a small team.

Formación

• Bachelor’s degree or higher in Computer Science, Mathematics, or related field or equivalent experience.
• Proficiency coding in Python or similar languages, plus ML tooling (JAX, PyTorch, Pandas).
• Experience with large language models, human-in-the-loop data annotation systems, or AI evaluation frameworks.
• Willingness to use AI tools with rigorous scrutiny and human review where needed.
• Track record of addressing open-ended research questions with rigor in academia or industry.

Responsabilidades

• Evaluate judge protocols by creating diverse and high-quality evaluations from datasets and literature.
• Research and implement strategies to improve judge performance, including learning-to-defer and human-in-the-loop methods.
• Build tools and infrastructure to scale the research agenda, including large-scale inference and a public leaderboard.
• Lead or contribute to research and mentorship programs as needed.

Conocimientos

Python

Machine learning

PyTorch

JAX

Pandas

LLM evaluation frameworks

Research methodologies

Educación

Bachelor's degree or higher in CS/Math or related field

Herramientas

JAX

PyTorch

Descripción del empleo

Sampura Research, an AI safety research non-profit based in London, seeks founding Members of Technical Staff to shape the direction of research and engineering and own end-to-end problems within our agenda.

You will lead evaluation design, implement methods to improve judge performance, and build scalable infrastructure for research, publications, and benchmark releases while collaborating with a small team.