Research Engineer, Post-training

medraai; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Medra is building Physical AI Scientists: robotic systems that work hand-in-hand with biopharma partners to enable scientific breakthroughs faster than ever before. In this role you will define post-training recipes for AI models, build data pipelines, and create evaluations that prove improvements in experimental workflows, all inside live lab environments.

You will collaborate with scientists and robotics engineers to shape the tech direction of a new ML team and push the boundaries of

Qualifications

• Experience building AI-driven workflows into real-world systems.
• Strong problem solving for debugging complex systems.
• Clear grasp of probability, statistics, and ML fundamentals.

Responsibilities

• Define post-training recipes for AI models—from problem selection to data collection and experiments.
• Build and own the post-training data pipelines integrating internal and public data.
• Create meaningful evaluations that show models improve scientific protocols and assay development.
• Develop agentic systems with context management and custom tool calls to surface new scientific insights.
• Collaborate with scientists, robotics engineers, and operation teams to bring reasoning into live lab loops.
• Shape the ML team's engineering culture and technical direction.

Skills

Python

PyTorch

JAX

LLMs

Reinforcement learning

Tools

TensorFlow

Ray

Job description

What We're Building:

If you want to shape the future of science, come build with us. Medra is building Physical AI Scientists: robotic systems that work hand-in-hand with leading biopharma partners to enable scientific breakthroughs faster than ever before.

Physical AI that can operate scientific instruments with human-level dexterity.

Scientific AI that can analyze results, reason about next steps, and close the loop autonomously.

We shipped our first production system over a year ago, recently raised a $52M Series A, and are opening one of the largest autonomous labs in the US. We're a small, ambitious team and you'd be joining early.

The Team:

• We’re a team of passionate, mission-driven engineers from companies like Tesla, Amazon, SpaceX, and Neuralink. We’re collaborative and love moving fast, both with our product and on team trips skiing or go-karting!

• As a team, we love nerding out about engineering and robotics — plus other topics like race cars or cooking. We like learning new things and then sharing our new knowledge with each other.

• Our team is opinionated and straightforward. We don’t mind intense discussions about design tradeoffs. If we have arguments or miscommunication, we resolve conflicts quickly and empathetically.

In this role, you will:

• Define post-training recipes for our AI models — from deciding which problem matters and how to measure it, to engineering large data collections, to running ML experiments, to integrating post-trained models into production workflows

• Build and own the post-training data pipelines integrating both internal data and public data

• Create meaningful and trustworthy evaluations that tell us whether our models are improving scientific protocols and assay development

• Develop agentic systems with context management and custom tool calls to surface new scientific insights about experimental design in real lab environments

• Work closely with scientists, robotics engineers, and operation teams to bring reasoning capabilities into live experimental loops for leading biopharma partners

• Shape the engineering culture and technical direction of a new machine learning team that's redefining how life science R&D gets done

Let's talk if you have:

• Practical experience building AI-driven workflows into the real world

• Strong problem solving skills for debugging complex systems

• A clear grasp of probability, statistics, and ML fundamentals

• Ability to own the post-training stack end-to-end: data pipelines, harnesses, RL environments, and agentic evaluations, even when things are loosely defined

• Proficiency in Python and familiarity with at least one deep learning framework (e.g., PyTorch, JAX)

• Experience with LLMs, post-training, reinforcement learning, or agentic systems