Achira

Machine Learning Research Engineer (MLRE) - Research

San Francisco Office, Remote. Posted 6 months ago

The role

You will advance foundation simulation models by implementing and experimenting with research ideas. You will scale ML systems, train and evaluate models, and engineer scientific prototypes into production.

What you will do

  • Design and run experiments to test hypotheses for foundation model development.
  • Engineer meaningful evals and metrics for rapid model iteration.
  • Design, build and maintain scalable, reproducible libraries for training, experimentation, evaluation, and simulation.
  • Implement model architectures from literature and in-house research.
  • Enable agent-driven research and workflows and maintain guardrails on agentic tooling.
  • Help prepare manuscripts, software artifacts, and datasets for public release.

Required qualifications

  • Strong software engineering fundamentals with experience building reproducible pipelines and writing documentation.
  • Track record of observable artifacts such as GitHub or papers showing work in ML or scientific computing libraries.
  • Solid working knowledge of PyTorch and JAX and the modern ML research stack.
  • Comfortable with HPC or large-scale compute environments.
  • Sufficient scientific depth to engage with research questions through prior industry experience or a PhD.

Nice to have

  • Experience with equivariant architectures, geometric deep learning, or GNNs (NequIP, MACE, SchNet, PaiNN, or similar).
  • Familiarity with generative modeling: diffusion models, flow matching, score-based methods.
  • Regular involvement in open-source ML or scientific computing libraries.
  • Experience building agent-driven research, active learning, and data curation pipelines.

About the company

Achira is building a team of scientists, ML researchers, and engineers to move beyond the beaten path in drug discovery. The company develops world models for the physical microcosm to make biology at the molecular level predictable and designable.