AI Research Scientist

Lexsi Labs is a frontier AI research lab focused on building safe, interpretable, and aligned AI systems as capabilities scale toward advanced autonomous intelligence.

Our research operates at the intersection of agent learning, alignment, and interpretability, with the goal of developing the scientific foundations required for trustworthy AI systems.

Key research areas include:

• Multi-agent reinforcement learning and agent learning systems for scalable coordination, exploration, and long-horizon learning.
• Continual learning and adaptive agents that can evolve safely over time without catastrophic forgetting.
• Alignment strategies and behavioral control methods including reward modeling, safety fine-tuning, pruning, and unlearning.
• Mechanistic interpretability and model transparency to understand internal representations and reasoning behavior in large models.
• Information-theoretic approaches including mutual information and causal analysis to study how knowledge and decisions form inside deep networks.
• Foundation models for structured and tabular domains aimed at high-stakes real-world applications.
• Open research infrastructure and tooling for alignment experimentation, RL training, and model evaluation.
• Open science and global collaboration, with research hubs in Mumbai and Paris and publications in leading venues such as ICML, ACL, IJCNN, and WWW.
  
  

Our goal is to advance the science required to ensure that future AI systems remain transparent, controllable, and aligned with human intent.

As an AI Research Scientist, you will work on fundamental research problems at the frontier of agent learning, reinforcement learning, interpretability, and alignment.

Your work will involve designing new algorithms, studying the internal behavior of large models, building research systems for experimentation, and contributing to open research that advances the understanding and safety of AI systems.

This role offers the opportunity to pursue ambitious research ideas while collaborating with a team building the next generation of alignment methods and agent learning systems.

• Conduct foundational research on multi-agent reinforcement learning, agent exploration, continual learning, and scalable alignment strategies.
• Design and evaluate algorithms that improve the learning efficiency, coordination, and controllability of large AI systems.
• Develop methods for interpreting and analyzing internal model behavior, including representation analysis, circuit discovery, and causal reasoning tools.
• Investigate alignment techniques such as reward modeling, safety fine-tuning, pruning, unlearning, and training-time behavioral control.
• Explore information-theoretic and causal approaches for understanding how models represent knowledge and make decisions.
• Build research infrastructure and experimental frameworks for RL training, alignment experiments, and large-scale evaluation.
• Publish research in leading venues and contribute to open-source tools and alignment research infrastructure.
  
  

• PhD or Masters or equivalent research experience in Machine Learning, Artificial Intelligence, Computer Science, Mathematics, Physics, or related fields.
• Strong background in reinforcement learning, multi-agent systems, large language models, interpretability, deep learning theory, or information-theoretic ML.
• Proven ability to conduct independent research and publish in top-tier AI conferences such as NeurIPS, ICML, ICLR, ACL, CVPR, or AAAI.
• Hands-on experience training and experimenting with models using frameworks such as PyTorch, JAX, or TensorFlow.
• Strong programming and experimentation skills in Python and modern ML research tooling.
• Demonstrated interest in AI alignment, safety, interpretability, or trustworthy AI systems.