Artificial Intelligence Researcher

Cerebro; San Francisco (CA); On-site; USD 200,000 - 300,000; Full time

Benefits offered by this job

Equity

Job summary

Cerebro, a San Francisco AI startup, is seeking a Machine Learning Researcher to own a research direction in LLM context optimization and interpretability. You will investigate how information is represented inside model contexts, determine which tokens influence outputs, and develop efficient context representations.

You will read papers, form hypotheses, train models from scratch, and run numerous experiments on substantial GPU infrastructure.

Qualifications

• At least two years of machine learning or research experience.
• Clear evidence that you have trained models from scratch.
• Strong understanding of transformers and modern model training.
• Ability to translate a paper or rough research idea into a real training run.
• A genuine research mindset centered on experimentation and learning.
• High agency and the ability to define your own research agenda.
• A preference for building and testing over lengthy planning or publication.

Responsibilities

• Design and run experiments across context compression and LLM interpretability.
• Train models from scratch, including data collection, architecture design, training loops and evaluation.
• Explore new methods for representing model context more efficiently.
• Run large scale training experiments using latest generation GPU infrastructure.
• Build evaluation systems that measure the impact of compression on accuracy, latency and model behaviour.
• Reproduce relevant research and turn promising ideas into working models.
• Iterate rapidly across new architectures and training methods.
• Ship successful research directly into the core product.

Skills

ML research

Transformers

Experimentation

Model training

Research mindset

High agency

Independent work

Education

2+ years ML research experience

Tools

NVIDIA GPUs

Job description

$200,000 to $300,000 base salary plus equity

Most AI companies are focused on building larger models.

This team is solving a different problem: how to make every token count.

We are working with a recently founded San Francisco AI startup building proprietary machine learning models that analyse LLM context before it reaches the underlying model. Its technology identifies what information matters, removes what does not and enables customers to reduce inference costs and latency while preserving output quality.

The company has secured significant seed funding from leading Silicon Valley investors and founders of several category defining technology businesses. It has also achieved exceptional customer adoption within its first year.

The role

As a Machine Learning Researcher, you will own a research direction within LLM context optimisation and interpretability.

You will investigate how information is represented inside model contexts, determine which tokens meaningfully influence outputs and develop new models that represent context more efficiently.

This is applied research with an immediate path into production. You will read papers, form hypotheses, train models from scratch and run a high volume of experiments using substantial GPU infrastructure. Successful ideas will become part of a product used by real customers.

Every researcher owns their work from initial hypothesis through training, evaluation and deployment.

What you will own

• Design and run experiments across context compression and LLM interpretability
• Train models from scratch, including data collection, architecture design, training loops and evaluation
• Explore new methods for representing model context more efficiently
• Run large scale training experiments using latest generation GPU infrastructure
• Build evaluation systems that measure the impact of compression on accuracy, latency and model behaviour
• Reproduce relevant research and turn promising ideas into working models
• Iterate rapidly across new architectures and training methods
• Ship successful research directly into the core product

What we are looking for

• At least two years of relevant machine learning or research experience
• Clear evidence that you have personally trained models from scratch
• Strong understanding of transformers and modern model training
• The ability to translate a paper or rough research idea into a real training run
• A genuine research mindset centred on experimentation and learning
• High agency and the ability to define your own research agenda
• A preference for building and testing over lengthy planning or publication for its own sake

This is unlikely to be the right role if your experience is primarily limited to RAG systems, chatbot applications, API integrations or fine tuning existing models.

Particularly relevant backgrounds

• Pretraining, post training or reinforcement learning research
• Mechanistic interpretability or representation learning
• Novel transformer architectures or training methods
• Research experience within a leading university or frontier AI laboratory
• Technical leadership or founding experience within an early stage startup
• Exceptional achievement through Kaggle, IOI, ISEF, ICPC or similar competitions
• A strong academic background in applied mathematics, computer science, physics or engineering

Working environment

This is a highly ambitious, fully in person startup environment in San Francisco.

Researchers receive significant autonomy, access to substantial compute and the opportunity to influence both the research agenda and the direction of the company.

The role will suit someone who wants to work on a technically difficult and commercially important problem with meaningful ownership and upside.

Compensation

$150,000 to $300,000 base salary

Meaningful equity

Applicants should include evidence of models they have trained, research they have completed or technically difficult systems they have built.