As a Research Engineer (RE) at Relation, I work primarily as an embedded engineer within a specific product or research team, where the main focus is accelerating delivery, unblocking infrastructure, and raising the quality bar for experimentation. Embedding durations vary—from short, goal-driven engagements to long-term (>1 year) partnerships—allowing me to develop deep team expertise while remaining responsive to evolving business needs.
Alongside this core embedding, I dedicate a smaller portion of time (~20%) to cross-cutting initiatives that improve consistency and collaboration across ML teams. I act as the engineering point of contact for my embedded team, regularly syncing with other REs to avoid silos, align standards, and turn successful team-level experiments into shared company practices. This balance enables strong team ownership while still driving org-wide progress.
A major part of my role is bridging research and engineering cultures. I help define shared workflows, tooling, and quality standards across data, model development, tracking, and evaluation—so researchers can move faster without sacrificing reproducibility or robustness. This includes templated pipelines, standardized repo structures, strong testing practices, and clear design decision records to make projects easier to transfer and scale.
On the research side, I contribute directly to novel ML work in genomics, including transformer alternatives for DNA language models. I am a co-author of PatchDNA: A Flexible and Biologically-Informed Alternative to Tokenization for DNA (ICLR), and I work on optimizing large-scale DNA-to-function models for Variant-to-Gene (V2G) mapping—pairing scientific innovation with production-ready engineering.
Finally, I collaborate closely with platform engineering to ensure ML infrastructure is scalable, observable, and reusable. This includes optimizing workflows, benchmarking new hardware (e.g. NVIDIA DGX B200), standardizing dependency management, and promoting best practices for data versioning, model release management, experiment tracking, and centralized evaluation—so teams can switch projects seamlessly and focus on impact.