Agent Harness Engineer

About the role

Axiom is building an agentic infrastructure to replace animal testing and transform how pharmaceutical companies conduct drug discovery. You will develop the core scaffolding that allows frontier models to perform complex, long-horizon scientific analysis on proprietary human datasets.

Key facts

What you'll do

• Develop the infrastructure and tooling required to transform frontier models into autonomous scientific agents.
• Create sandboxed, deterministic execution environments that support reliable evaluations and reinforcement learning.
• Build and maintain data pipelines for agent trajectories, runtime context, and training data.
• Design evaluation systems, including offline test suites, regression gates, and LLM-as-judge pipelines.
• Implement observability and debugging tools to track model calls, tool usage, and state transitions.
• Engineer memory, retrieval, and recovery systems to ensure agent coherence during extended tasks.
• Collaborate with domain experts to translate scientific standards into measurable rubrics and golden sets.
• Manage the agent loop, including guardrails, output verification, and resource budgeting.
• Support ML research by building reward instrumentation and rollout infrastructure.

Requirements

• Proven experience building and shipping agentic systems using LLM APIs, including tool use and control loops.
• Strong software engineering background with a focus on infrastructure, platform, data, or devtools.
• History of building custom evaluation, monitoring, or RL environment observability tooling.
• Ability to own outcomes from start to finish without waiting for detailed specifications.
• Preference for simple, maintainable system design over unnecessary complexity.
• Experience reading and debugging complex agent failure traces.

Skills & tools

• Python
• Modal
• DuckDB
• FastAPI
• Docker
• Containerization
• Terraform
• SvelteKit
• Svelte 5
• React

Practical notes

• You will be working in an evolving field where you must define the playbook as you build.
• We prioritize engineers who maintain a high degree of curiosity and a hands-on tinkering mindset.
• Success in this role requires a focus on reliability to prevent silent environment failures from corrupting training data.