Agent Harness Engineer

Responsibilities

• Own the agent harness, including scaffolding, tooling, and infrastructure for long-horizon scientific analysis
• Build data pipelines, storage, and systems for runtime context, agent trajectories, and training data
• Develop sandboxed execution environments for reproducible and deterministic evals and RL
• Design and run evaluation systems, including offline suites and LLM-as-judge pipelines
• Encode domain expert knowledge into measurable rubrics, golden sets, and review workflows
• Build observability and replayability tools to trace model calls, tool calls, and state transitions
• Engineer context management including memory, compaction, retrieval, and recovery
• Manage the agent loop, including retries, budget caps, stop conditions, and guardrails
• Support ML research with environments and rollout infrastructure for RL

Requirements

• Proficiency in Python, Modal, DuckDB, FastAPI, Docker, and Terraform
• Experience building with LLM APIs and shipping agentic systems (tool use, loops, debugging)
• Experience building bespoke evaluation, monitoring, or RL environment observability tooling
• Experience with SvelteKit, Svelte 5, or React
• Strong software engineering foundation with depth in infrastructure, platform, or data systems
• Ability to own outcomes end-to-end and ship clean, maintainable code

About the Company

Axiom is building a compounding ecosystem to replace animal testing and reshape clinical trials. By combining world-class datasets with an agentic harness, Axiom enables frontier models to reason over biological data to predict human drug outcomes more accurately than animal models.