Backend Software Engineer (Research team)

Apollo Research is a London-based AI safety organisation focused on high-risk failure modes of frontier AI models, particularly loss of control and deceptive alignment in language models. We specialise in LLM agent evaluations for strategic deception, evaluation awareness, and scheming.

As a Backend Software Engineer on the Research team, you will develop tools supporting frontier AI safety research. You'll build evaluation libraries and systems for monitoring LLM traffic. Our entire stack is Python. We have switched to Inspect as our primary evals framework. You will build LM agent evaluations, scaffolding, and tooling used to evaluate frontier models from labs such as OpenAI, Anthropic, and Google DeepMind.

Potential first-year projects:

• Analysis tools for investigating thousands of evaluation runs simultaneously
• Automated pipelines for pre-deployment model testing
• Distributed evaluation orchestration with security features
• Real-time LLM proxy monitoring systems
• CI/CD pipeline improvements
• Data warehousing for evaluation transcripts
• Inspect framework ecosystem enhancements

Essential: production-quality Python code experience.

Valued background examples:

• Led software tool development over 1+ years
• Built startup tech stacks
• Authored/maintained popular open-source libraries
• Programming competition placements (IOI, ICPC)
• 5+ years professional software engineering

Bonuses: LLM agent/evaluation experience, infosecurity background, AWS expertise, AI safety interest.

Location: London (primary) or San Francisco office. In-person role.
Salary: £100,000 - £200,000 GBP.
Benefits: flexible hours, unlimited vacation, provided meals, paid work trips, $1,000 professional development budget. Supported by AI Futures Grants (up to £10,000 relocation/visa reimbursement).