Research Scientist

Key Talent Solutions; San Francisco (CA); On-site; USD 150,000 - 250,000; Full time

Job summary

Key Talent Solutions seeks a Research Scientist - Post Training to advance how training data shapes LLM capabilities, generalisation and alignment. This hands-on role blends research, evaluation and collaboration with external partners.

You will design SFT and RL experiments, build reproducible benchmarks, publish results, and translate findings into practical improvements for data quality and production. A founding-level background from a top technical university is preferred.

Qualifications

• Must have held a founding-level role
• Graduated from a top-tier technical university
• Strong familiarity with LLM training and evaluation methodologies
• Ability to design lightweight experiments and extract actionable insights
• Experience at a frontier AI research lab

Responsibilities

• Design and run controlled SFT and RL experiments to measure the impact of training datasets on model performance
• Evaluate how different data sources affect model capabilities, generalisation and alignment
• Help develop public evaluations, benchmarks and new data types that advance LLM capabilities
• Publish external-facing research, technical reports and blog posts
• Collaborate with internal stakeholders to improve data quality based on experimental findings
• Analyse complex or noisy experimental results and turn them into clear, actionable insights
• Contribute to reproducible evaluation methods across models, capabilities and technical domains

Skills

LLM training

Evaluation methodologies

Experiment design

Cross-domain collaboration

Hands-on builder

Frontier AI experience

Education

Top-tier technical university

Job description

A Series B AI data company, valued at about $3.2B, that builds expert datasets, evals and RL environments for frontier AI labs.

About the Role

We are looking for a Research Scientist - Post Training to join a post-training research team focused on understanding how training data improves the capabilities, generalisation and alignment of large language models (LLMs).

In this role, you will design and run controlled training experiments using supervised fine-tuning (SFT) and reinforcement learning (RL)-based post-training techniques. You will investigate how different datasets influence model behaviour and translate your findings into actionable recommendations for improving data quality, selection and production.

Working closely with research partners, you will generate clear, reproducible evidence of how specific datasets contribute to measurable model improvements. You will also contribute to evaluation infrastructure and benchmarks across models, capabilities and domains.

This is a hands-on, experimental role combining research, technical evaluation and collaboration with external partners.

Key Responsibilities

• Design and run controlled SFT and RL experiments to measure the impact of training datasets on model performance.
• Evaluate how different data sources affect model capabilities, generalisation and alignment.
• Help develop public evaluations, benchmarks and new data types that advance LLM capabilities.
• Publish external-facing research, technical reports and blog posts.
• Collaborate with internal stakeholders to improve data quality based on experimental findings.
• Analyse complex or noisy experimental results and turn them into clear, actionable insights.
• Contribute to reproducible evaluation methods across models, capabilities and technical domains.

Required Qualifications

• Must have held a founding-level role and graduated from a top-tier technical university
• Strong familiarity with LLM training and evaluation methodologies.
• Ability to design lightweight experiments, work quickly and extract actionable insights from complex results.
• Comfort working across multiple domains, including finance, software engineering and policy.
• A hands-on, building-oriented mindset, with a focus on experimentation and practical outcomes.
• Previous experience at a frontier AI research lab.

Red Flags

• PhD-only researcher profile with no demonstrated track record of shipping practical solutions. The role explicitly favours hands‑on builders over purely academic researchers.
• Treats evaluation as reporting benchmark scores rather than diagnosing why model behaviour changed.

Green Flags

• Has built evaluation harnesses, training pipelines, benchmarks or other experimental infrastructure.
• Has worked at or alongside a frontier AI lab, RL environment company, AI safety organisation or benchmarking organisation.
• Has published research or completed meaningful projects involving model evaluation, post‑training or data curation.

Compensation and Benefits

• Base salary of $150,000–$250,000 per year.
• Additional compensation through profit sharing, with total cash compensation potentially reaching $250,000–$450,000.