Research Engineer

Key Talent Solutions; San Francisco (CA); On-site; USD 180,000 - 350,000; Full time

Job summary

Key Talent Solutions partners with AI ventures building research, evaluation, and data infrastructure for frontier-model teams. We seek research-minded engineers who have shipped RL environments, eval harnesses, reward and post-training pipelines, reliable enough to gate real releases.

This role focuses on RL environments, evaluation infrastructure, and production-grade data systems that shape training through evaluation and signals.

Qualifications

• Must have shipped RL-related work including environments, evaluation harnesses, reward systems or post-training pipelines.
• Experience turning research into production-grade systems for model evaluation and training.
• Strong publication or production-ready code in related frontier-model work is highly valued.

Responsibilities

• Own RL environments and evaluation infrastructure against real targets.
• Develop post-training reward-signal and data pipelines feeding model improvement.
• Build systems that translate production signals into evaluation specifications for training.

Skills

RL environments

Model evaluation

Data pipelines

Research-to-production

Job description

Build the evaluation and RL infrastructure that shapes frontier models.

Key Talent Solutions partners with a portfolio of venture-backed AI companies building the research, evaluation, and data infrastructure that frontier-model teams depend on. We are looking for research-minded engineers who have actually shipped this work: RL environments, eval harnesses, reward and post-training pipelines, reliable enough to gate real releases.

What you'd own

• RL environments and evaluation infrastructure against real, not synthetic, targets
• Post-training, reward-signal, and data pipelines that feed model improvement
• The systems that turn messy production signals into evaluation specs that shape training

What we're looking for

• Strong research grounding paired with real shipping. Seats range from exceptional early-career builders to senior research engineers
• Genuine depth in RL environments, model evaluation, fine-tuning, or interpretability, not just applied prompting
• A track record that reads as elite: top-tier lab or company, top competition results, or published research, alongside code that runs in production
• This is a high bar. Applied-only profiles with no research depth are not a fit

Strong bonus signals

• Eval, RL, or data-infrastructure work in the frontier lane (the METR, Mechanize, Scale, Surge lane)
• Publications at top conferences, or an open-source agent or eval framework you authored
• Computer-vision research depth for the applied-research seats

Compensation: $180K to $350K+ base, plus profit share or equity depending on the client. Exact range depends on stage and level.

Not a fit: research-only profiles that have never shipped, and applied engineers with no real evaluation or research depth.