AI Research Engineer

Straker Ltd•Philadelphia, PA

4d•Hybrid

About The Position

AI Research Engineer Do research that ships, and build the systems around it. arbitr isn't asking you to fine-tune someone else's roadmap. We're asking you to help define it. We build models that run inside our platforms. We build the retrieval, memory, and evaluation systems around them, and we ship the whole thing into production, where it has to actually work. arbitr is a 25-year global company (formerly Straker Ltd) — ASX-listed, offices across the USA, Japan, Europe and New Zealand - that has always been at the forefront of technology, and that includes AI , from the inside out. We operate in smaller teams, as a startup, which means you get something rare: the scale and stability of an established business, with the mandate and the appetite for risk of a team that is genuinely starting again. If you want to do research that ships, not research that lives in a notebook, this is that role.

Requirements

Trained a model from the ground up, not just fine-tuning.
Understand what's happening under the hood and can debug when it goes wrong.
Think in experiments and build real evaluation gates before trusting results.
Relentless about getting from idea to production.
Filter hard for signal versus noise.
Detailed understanding of model building concepts like LoRA, PT, reinforcement learning, and inference.
Comfort designing and building datasets.
Working background in data science.
Ability to work alone on a project or with a team.
Strong Python skills.
Self-starter mindset; personally accountable for outcomes.
Familiarity with the ideas at labs.straker.ai or the appetite to get familiar quickly.
Must have working rights for USA.

Responsibilities

Build custom models for customers and across arbitr's platforms, owning the pipeline end-to-end from first experiments through production and beyond.
Build the systems around the models, including retrieval, memory, reasoning, and orchestration, to deliver value in production.
Own our evals by building benchmark tooling to assess model performance.
Push the research by tracking new models and training techniques and integrating relevant ones into our stack.
Design research workstreams, taking ideas from problem statements through experiments, findings, and pipeline changes.
Deploy and run models at scale on GPU infrastructure, including serving, monitoring, and orchestration.
Work at the partner table, supporting strategic engagements with global partners.
Go wide across arbitr by providing engineering support to wider platform work, including architecture reviews, AI strategy input, and hands-on development.
Provide technical guidance to product, engineering, and partnerships as AI experts.