Intrinsic Safety
Research Engineer, Judgment Systems
3mo ago
250000 –400000 USD / yearUSAmodel evaluationmachine learningexperimental designresearchmodel trainingfine-tuning
Research Engineer role to design evaluations, study failures, and build research loops to improve AI judgment systems at Intrinsic Safety in San Francisco.
Nice to have
- Experience training, fine-tuning, or evaluating modern ML systems
- Strong programming skills and comfort working in research-heavy codebases
- Familiarity with LLMs, agent systems, post-training, reinforcement learning, retrieval, or adjacent areas
- Ability to design clean experiments and draw reliable conclusions from noisy results
- Strong engineering judgment and a bias toward building
- Interest in fraud, risk, trust and safety, compliance, or other regulated and adversarial domains
Conditions
- Competitive salary and meaningful equity
- Platinum-level medical, dental, and vision insurance
- Unlimited PTO, sick leave, and parental leave
- Up to $100 per month in reimbursement for personal health and wellness expenses
- 401(k) plan
Other
- At Variance, we are teaching machines to make the hardest judgment calls at scale. We build AI agents for the high-precision gray area of stopping fraud, scams, and abuse. This isn't another sales tool or a customer service system. We're solving real problems in investigations and fraud prevention to protect innocent people from being harmed.
- We’re a small, talent-dense team in San Francisco working on a problem at the edge of what AI systems can reliably do: making good decisions in messy, adversarial, real-world environments.
- We’re looking for a Research Engineer to help push that frontier forward. You’ll design evals, study failures, build new research loops, and turn research ideas into production capabilities.
- This role sits at the intersection of research and engineering: part model builder, part experimentalist, part systems engineer.
- Care deeply about protecting people from fraud, scams, and abuse
- Have strong opinions about model quality, evaluation, and experimental rigor
- Want to work on core model and agent behavior
- Are excited to train, fine-tune, and improve models for hard real-world judgment tasks
- Think in tight research loops: hypothesis, experiment, evaluation, failure analysis, iteration
- Thrive in ambiguous, fast-moving environments where the path is not obvious and the feedback loop is short
- Are motivated by the challenge of making AI systems work in adversarial, regulated, and high-consequence settings
- Want to help define what trustworthy AI means in real-world use cases
- Train, fine-tune, and improve models for fraud, scams, abuse, and other high-stakes judgment workflows
- Own research threads focused on improving agent capability, reliability, and decision quality
- Build proprietary benchmarks, datasets, and evals that reflect real customer workflows, regulatory constraints, and real failure modes
- Design and run experiments across post-training, retrieval, tool use, planning, memory, and long-horizon agent behavior
- Study where models break, why they break, and how to make them more robust
- Prototype new training strategies, agent architectures, and evaluation methods, then turn the best ideas into production systems
- Work closely with founders and engineering to translate research advances into deployed product capabilities
- Push the boundary of what AI agents can do in regulated industries
- Our models get materially better at making hard judgment calls in production
- Our models are trusted at scale
- We develop evals and training loops that compound over time
- We understand failure modes more clearly and improve system behavior faster
- New research ideas turn into real product capabilities quickly
- We believe in ownership, urgency, and craft. We enjoy spirited debate, wild ideas, and building things we’re proud of. We’re fully in-person in San Francisco.