Who we are We build AI systems that remember, perceive, reason and adapt over time. A big part of our research is getting perception and memory to work together. The role You'll work with our research team on computer vision, representation learning, and how what a system sees turns into what it remembers. Right now that includes HyperGNN encoders, JEPA-style predictive learning, self-supervised learning and Bayesian methods for handling uncertainty. There's also room to work on VLA models and real sensors or robots. Your job is to take ideas, build them, and find out honestly whether they work.

What you'll do
• Build and evaluate perception models for images, video and multimodal data
• Try predictive and self-supervised approaches, including JEPA-style models
• Use Bayesian and probabilistic methods when data is noisy or incomplete
• Connect perception to our memory systems so they keep track of relationships, events and change over time
• Work with VLA models and integrate sensors or robots when a project needs it
• Reproduce papers, set baselines, and design experiments that show where an approach breaks
• Write clean research code and document results so others can build on them What we're looking for
• 2–5 years of hands-on work in research, industry or substantial independent projects




• Strong Python and PyTorch, including training and evaluating models
• Real depth in at least one of: computer vision, representation learning, probabilistic ML, graph neural networks or embodied AI
• Solid ML fundamentals, and the ability to explain why a model behaves the way it does
• A track record of taking an idea from a paper to a working, evaluated implementation
• Work we can look at: papers, open-source contributions, projects or shipped systems
• Initiative, honesty about results, and clear explanations of your choices Nice to have
• JEPA, self-supervised learning or world models
• Bayesian learning or uncertainty estimation
• Graph-based or structured representations
• Video, temporal or multimodal models
• VLA models, robot learning or hardware integration
• Deep expertise in one area matters more than covering all of them. Why join You'll own real research projects, have room to try unusual ideas, and be expected to test them properly. We care about explicit thinking, good code and results that hold up.

How to apply
Fill in this form. We'll ask for your CV, links to your best work and a short write-up of one or two projects -> https://binary.so/5vjppCB

📌 Research Engineering Intern (Bengaluru)
🏢 MetaCognition
📍 Bengaluru