← All roles
Engineering
Agent Training Infra Engineer
This is not a traditional full-stack role. We need someone who can build systems, platforms, and tools, and also understands agent training, model data, and task-environment construction. You don’t need to start as an RL expert, but you should have touched related systems or data pipelines and be willing to go deeper. Your core output is runnable, verifiable, scalable task environments and data-production systems.
What you’ll do
- 01Build the AI data-production pipeline — task creation, data collection, human / expert and AI-assisted labeling, QC, task routing, and delivery.
- 02Build the RL environments for agent training — abstracting real tasks into trainable, evaluable, reproducible interactive environments.
- 03Design mechanisms for state, action, reward, success criteria, environment reset, and trajectory logging.
- 04Own core system engineering — task management, labeling / review tools, environment runtime services, APIs, and workflow orchestration.
- 05Build quality mechanisms for data and environments — task authenticity, environment stability, answer verifiability, reward soundness, and data validity.
- 06Use AI coding tools and agent workflows deeply to raise the efficiency of data production, environment building, and delivery.
What we look for
- 01Bachelor’s or above in CS, software engineering, AI, or related.
- 02Solid engineering — strong depth on at least one of front-end / back-end, able to deliver independently on the other.
- 03Familiar with at least one mainstream stack: TypeScript / JavaScript, Python, Go, Node.js, or Java.
- 04Understanding of APIs, databases, task queues, caching, permissions, logging, and monitoring.
- 05Built at least one of: agent, RL environment, benchmark, eval, sandbox, automation platform, labeling platform, or data-production platform.
- 06Understanding of core RL / agent-training concepts: state, action, reward, trajectory, rollout, environment reset, success criteria.
- 07Good engineering habits — code quality, testing, maintainability, and system stability.
- 08Comfortable with early-startup ways of working — ambiguity, self-directed problem-solving, and driving results.
Nice to have
- Heavy user of AI coding tools (Claude Code, Cursor, Codex, etc.).
- Experience with MCP, agent workflows, browser automation, RPA, sandboxes, Docker, or task scheduling.
- Built systems from 0 to 1 — startups, open-source, or core engineering in small teams.
- Your own views and ongoing tracking of model training, post-training, agent eval, RL infra, or AI data pipelines.
Apply for this role
Send your résumé and a short note on why this role.