Research Engineer, Training Infrastructure

About Goodfire

Behind our name: Like fire, AI holds the potential for both immense benefit and significant risk. Just as mastering fire transformed human history, we believe the safe and intentional development of AI will shape the future of our species. Our goal is to tame this new fire.

Goodfire is an AI interpretability research company focused on understanding and intentionally designing advanced AI systems. We believe advances in interpretability will unlock the next frontier of safe and powerful foundation models and that deep research breakthroughs are necessary to make this possible.

Everything we do is in service of that mission. We move fast, take ownership, and constantly push to improve. We believe in acting today rather than tomorrow. We care deeply about the success of the organization and put the team above ourselves.

Goodfire is a public benefit corporation headquartered in San Francisco with a team of the world’s top interpretability researchers and engineers from organizations like OpenAI and DeepMind. We’ve raised $57M from investors like Menlo, Lightspeed and Anthropic and work with customers including Arc Institute, Mayo Clinic, and Rakuten.

The role:

We're seeking a research engineer to lead the development of our model training infrastructure. You'll own the critical systems that transform pre-trained models into safe, capable, and reliable AI systems through fine-tuning, RLVR, and other post-training techniques.

Key responsibilities:

• Design and implement scalable and customizable post-training pipelines (SFT, RLVR, DPO)
• Develop suitable evaluation frameworks
• Optimize inference-time interventions and model serving for post-trained models
• Collaborate with research teams to rapidly prototype and validate new techniques

Who you are:

Goodfire is looking for experienced individuals who embody our values and share our deep commitment to making interpretability accessible. We care deeply about building a team who shares our values:

Put mission and team first
All we do is in service of our mission. We trust each other, deeply care about the success of the organization, and choose to put our team above ourselves.

Improve constantly
We are constantly looking to improve every piece of the business. We proactively critique ourselves and others in a kind and thoughtful way that translates to practical improvements in the organization. We are pragmatic and consistently implement the obvious fixes that work.

Take ownership and initiative
There are no bystanders here. We proactively identify problems and take full responsibility over getting a strong result. We are self-driven, own our mistakes, and feel deep responsibility over what we’re building.

Action today
We have a small amount of time to do something incredibly hard and meaningful. The pace and intensity of the organization is high. If we can take action today or tomorrow, we will choose to do it today.

If you share our values and have at least two years of relevant experience, we encourage you to apply and join us in shaping the future of how we design AI systems.

What we are looking for:

Required experience:

• 4+ years of experience in ML engineering, with at least 2 years focused on LLMs or foundation models
• Deep expertise in fine-tuning, RLVR, and modern post-training techniques
• Production experience deploying and maintaining language models at scale
• Technical proficiency in Python, PyTorch/JAX, and distributed training frameworks
• Mission alignment with building safe and powerful AI systems

Core competencies

Post-training excellence

• Expert understanding of supervised fine-tuning, RLVR, DPO
• Experience with preference modeling and reward model training
• Hands-on experience with parameter-efficient fine-tuning (e.g., LoRA, QLoRA)

Infrastructure and scale

• Building systems that handle diverse training workflows efficiently
• Optimizing training for large models for compute efficiency
• Multi-node distributed training

Accelerating research

• Rapid prototyping of novel post-training techniques
• Building flexible infrastructure that supports multiple research directions
• Quickly adaptiing to new research ideas

Preferred qualifications:

• Experience with RL and policy gradient methods
• Published work on model alignment and/ or post-training techniques
• Familiarity with interpretability tools and research on the mechanistic understanding of model behavior

Compensation & benefits:

This role offers market competitive salary, equity, and competitive benefits. More importantly, you'll have the opportunity to work on groundbreaking technology with a world-class team on the critical path to ensuring a safe and beneficial future for humanity.

The expected salary range for this position is $180,000-350,000 USD.