Skip to main content

Research at Primateria

Job Description

More details: https://jobs.primateria.com/

About the role

This is a six-week long paid , with the option to extend to more weeks based on your work with us.

Your role is to perform a scoped research and engineering project related to solving the multi-referential Schrodinger’s equation that could be open-sourced with a written report or a publication in an academic setting, such as a conference paper.

Our current approach is to extend the architecture of models such as the Orbformer with physics-inspired features such as adding coupled clusters, correction predictions to CASSCF-like determinants, and periodicity. Potential ideas for you to work on include:

  1. Transition-metal benchmark suite: Build a rigorous benchmark suite for 3d atoms and small transition-metal oxides, including same-Hamiltonian references, spin states, and VMC error decomposition.
  2. Better nodes for DMC: Use Orbformer or another neural wavefunction model as a trial wavefunction for DMC and measure whether the learned multi-reference structure improves fixed-node error.
  3. Amortized active-space conditioning: Condition the model on active-space metadata, orbital occupations, nuclear charge, and geometry so it can transfer across related multi-reference systems.

Feel free to suggest more ideas to us!

Resources

We have a compute cluster of 66 RTX Pro 6000 GPUs that we purchased to efficiently run multiple Monte Carlo walkers in parallel; this full cluster is expected to be available on June 25, 2026. Before June 25, you will have access to our current 10 RTX Pro 6000 GPUs. Our setup is customizable in that you could use either all GPUs at once or just use a single machine with 8 GPUs. You will have unlimited access to all of our machines, including physical access at our colocation data center.

If you need more powerful GPUs such as a DGX B200 for good reasons (e.g. you want to start training a real foundation model to test hypotheses that you verified with our existing cluster), we can always rent them.

You will also have unlimited tokens from any foundation model provider you prefer (e.g. Anthropic and OpenAI) to assist in your research and engineering work.

Who you are

  1. You are currently in a graduate degree program or hold a graduate degree in physics, chemistry, computational chemistry, applied mathematics, electrical engineering, computer science, or a related field.
  2. You would like to solve the particular problem of solving the multi-referential Schrodinger’s equation and have many ideas on how to get there.
  3. You have exciting ideas that you want to execute on; we will give you all the compute and tokens you need to shine.
  4. You have experience working with machine learning and quantum chemistry libraries such as JAX and PySCF.
  5. You are comfortable reading machine learning papers and modifying machine learning research code.
  6. You have the ability to work independently in a field with ambiguous approaches to a common research problem.
  7. You have the ability to execute on research ideas and can communicate clearly about experiments, failures, and where we can help for next steps.
  8. You are collaborative and open to sharing research approaches with everyone on the team. You are flexible and consider all approaches towards tackling a shared problem with your team.

Compensation

The base compensation for this role is $20,000 for six weeks of . The same pay will be pro-rated if the is extended further.

Location

We would strongly prefer to work in person in Palo Alto, though we are flexible right now since we don’t have a formal company office yet. We currently work in Palo Alto from co-working spaces.