Machine Learning Scientist I/II, Multi-Modal Scientific Reasonings

Your Impact at LILA

We’re hiring a Machine Learning Scientist to advance multi‑modal reasoning with vision‑language models (VLMs) on real-world scientific data including, but not limited to: figures and plots, microscopy data from diverse sources. You’ll design and build state‑of‑the‑art methods to advance the state of Scientific Superintelligence.

What You'll Be Building

• Lead research on multi‑modal reasoning systems that interpret scientific data (images, plots, text, etc) using state‑of‑the‑art and custom VLMs.
• Design training, adaptation and test-time methods and strategies (e.g., instruction tuning, supervised learning, RLHF, RAG) for scientific understanding tasks.
• Build datasets and benchmarks from real scientific artifacts (e.g., microscopy, spectra, protocols) to understand model performance.
• Develop perception modules (e.g, OCR, table/structure recognition, plot parsing) for multi-modal data modalities.
• Collaborate with domain scientists and engineers to scale research into production ready systems for scientific superintelligence.

What You’ll Need to Succeed

• Advanced degree in a relevant field (CS/AI, Applied Math/Stats, EE) or a physical‑sciences discipline (Materials, Chemistry, Physics) with strong ML focus; or equivalent research/industry experience.
• Track record in multi‑modal ML or VLMs demonstrated via shipped systems, publications, or open‑source.
• Understanding of scientific QA/benchmarks and custom evaluation design.
• Experience with multi-modal fine-tuning, document parsing & understanding, dataset curation and benchmarking.
• Strong engineering skills centered on modern machine learning frameworks (e.g., PyTorch, Huggingface).
• Clear communication and collaboration in cross‑functional settings.

Bonus Points For

• Experience with scientific data modalities in real-world laboratories such as microscopy images.
• Publications in top ML/CV/NLP venues or tangible impact in applied industrial research.
• Contributions to open‑source multi‑modal tooling, evaluation suites, or datasets.