Founding AI Scientist: Multimodal, Continual-Learning Lead

Next Phase Recruitment; New York (NY); On-site; USD 150,000 - 190,000; Full time

Job summary

Next Phase Recruitment in New York invites applications for a senior AI researcher focused on foundational agent design, continual learning, and multimodal reasoning. You will work with a world-class team, shape long-horizon research directions, and drive projects from concept through evaluation on real-world scientific data.

The role emphasizes publications in NeurIPS/ICML-like venues, developing embodied AI systems that learn with minimal retraining, and delivering metrics that capture

Qualifications

• PhD or equivalent in CS/AI/ML or related field.
• Publications at NeurIPS/ICML/ICLR/CoRL/CVPR.
• Experience across at least two domains among vision, language, reinforcement learning, generative models, or robotics.

Responsibilities

• Join as a key contributor shaping foundational research in intelligent agent design.
• Lead research into continual learning and memory systems without retraining.
• Develop multimodal learning methods for real-world, partially observed data.
• Create embodied agents that act, reason, and learn in real time.
• Design evaluation metrics emphasizing adaptation and competence.
• Publish in leading conferences and help define the next frontier in AI for science.

Skills

Reinforcement learning

Continual learning

Reward modeling

Preference learning

Data-driven experimentation

Multimodal learning

Embodied agents

PyTorch

TensorFlow

Education

PhD or equivalent experience in Computer Science, AI, ML, or related fields

Tools

PyTorch

TensorFlow

Job description

Next Phase Recruitment in New York invites applications for a senior AI researcher focused on foundational agent design, continual learning, and multimodal reasoning. You will work with a world-class team, shape long-horizon research directions, and drive projects from concept through evaluation on real-world scientific data.

The role emphasizes publications in NeurIPS/ICML-like venues, developing embodied AI systems that learn with minimal retraining, and delivering metrics that capture