AI Agent Engineer (Autonomous Driving)

The role

This person will develop agent systems using LLM/VLM-based agents and tool calling to automate the evaluation cycle of autonomous driving models. They will design, experiment, and integrate evaluation pipelines to improve driving performance.

What you will do

• Design agent reasoning loops and execution structures so VLM can select and call tools to query driving videos, simulation logs, and evaluation metrics.
• Develop agent workflows that perform multi-step driving evaluation and failure triage through context management, tool routing, and error recovery.
• Generate and improve the quality of structured pseudo-labels for driving decisions and estimated causal factors based on agent evidence.
• Verify agent judgment and label quality using human annotation and counterfactual scenarios and integrate workflows into the closed-loop evaluation pipeline.

Required qualifications

• 3 or more years of experience in Software Engineering, Machine Learning, or AI application or equivalent research/development capabilities.
• Experience designing and implementing Python-based AI applications or data pipelines.
• Experience developing and evaluating execution results of LLM/VLM-based systems including tool calling or multi-step workflows.
• Ability to design evaluation metrics and datasets to verify agent judgment and label quality and analyze failure cases.
• English communication skills to discuss and collaborate on technical issues with overseas offices.

Nice to have

• Experience in scene understanding and failure analysis using autonomous driving data or closed-loop simulation.
• Experience in VLM fine-tuning, auto-labeling, or causal reasoning.
• Experience in agent workflow development using long-term memory/retrieval, custom tool development, or multi-agent orchestration.
• Experience in analyzing and improving the reliability and bias of LLM/VLM-based evaluators.

Before you apply

• A 3-month probation period may apply.
• Employment may be canceled if false information is found in the application.
• Reference checks may be conducted after the interview process with the applicant's consent.

About the company

42dot's autonomous driving VLA team develops technology that enables vehicles to understand complex road situations and connect judgments to driving actions through Vision-Language-Action (VLA) models.