Your mission
Invent the representations that make matter legible to machines
Most AI systems succeed because the world they model is forgiving. Materials discovery is not.
At Dunia, we are building AI systems that must reason about molecules, materials, and physical processes across scales, regimes, and sparse data. The limiting factor is not model size. It is representation.
As ML Researcher, Representation Learning, you work on the hardest layer of AI for Materials: deciding what the machine should see. You invent representations that let models generalize, transfer, and reason under uncertainty, and you test those ideas inside real discovery loops that include experiments and simulations.
This role is for someone who wants their representation ideas to matter beyond papers.
Your tasks will include:
Define how matter is encoded - Design representations for molecules, materials, and processes that respect physical and chemical structure
- Decide what information should be explicit, implicit, or learned
- Build abstractions that transfer across tasks and domains
Build multi-modal foundation models - Combine graphs, sequences, text, images, simulations, and experimental data
- Explore how different modalities reinforce or contradict each other
- Move beyond benchmark metrics toward models that support real decisions
Work where research meets reality - Test ideas in production-grade systems used by scientists
- Collaborate closely with chemists, physicists, and automation teams
- Learn quickly which ideas hold up and which don’t
Push representation learning forward - Introduce probabilistic reasoning and uncertainty awareness
- Challenge existing paradigms when they break down in physical domains
- Help shape the broader direction of AI-native science at Dunia
Your profile
- PhD or equivalent training in machine learning, computer science, or a closely related quantitative field, with4–8 years of post-degree experience in research-driven ML roles
- Deep experience in representation learning and modern deep learning architectures
- Track recordof designing representations for structured or scientific data that generalize beyond a single task or dataset
- Strong software engineering skills, with experience turning research ideas into robust, maintainable systems
- Comfortable working at the boundary ofresearch and production, theory and engineering
- Curious about the physical world and motivated by applying ML to chemistry, physics, or materials
- Willing to challenge existing paradigms when they break down in real-world settings
- Fluent in English; additional language desirable
