Agentic AI Expert

At VinDynamics, we are building the next generation of humanoid robots with real cognitive intelligence — robots that don't just talk, but can reason, plan, remember, and act autonomously in complex real-world environments.

We believe the future of robotics is Agentic AI: systems that can set goals, decompose tasks, use tools, learn from experience, and adapt safely over time.

To make that vision real, we are looking for an Agentic AI Expert to help design the cognitive brain of our humanoid robots.

About the role

As an Agentic AI Expert, you will work at the intersection of LLMs, multimodal perception, memory, and robot action.

Your mission is to define how robots think, not just what they say.

This role goes far beyond conversational AI — you will design goal-driven, embodied intelligence that connects reasoning, perception, and control into a unified cognitive system.

What you'll do

• Design and evolve agentic cognitive architectures for humanoid robots
• Build LLM-based reasoning, planning, and decision-making loops (Planner–Executor, ReAct-style agents, tool-use)
• Develop robot memory systems (short-term, long-term, episodic, semantic)
• Ground language ↔ vision ↔ robot state ↔ physical action
• Define evaluation metrics for autonomy, robustness, safety, and latency
• Collaborate closely with Vision AI, Conversation AI, Control, and System teams
• Translate cutting-edge research ideas into production-ready robot intelligence

What we're looking for

• Master's or PhD in AI, Machine Learning, Robotics, or related fields
• Strong hands-on experience with LLMs, agent architectures, and decision-making systems
• Solid Python skills, with PyTorch as a core framework
• Strong system-thinking mindset and experimental rigor

Bonus if you have experience with:

• Agentic AI systems (Planner–Executor, ReAct, multi-agent setups)
• Memory-augmented agents (RAG, long-term memory, episodic memory)
• Multimodal agents (vision + language + action)
• Reinforcement Learning, RLHF, preference learning
• Safety, alignment, and failure recovery for autonomous agents
• Research publications (NeurIPS, ICLR, ICML, ICRA, RSS)