About Distyl AI

Distyl is an applied AI technology company partnering with the world’s most ambitious institutions to rearchitect critical operations for the frontier of AI. Our customers include the largest companies in telecom, healthcare, insurance, manufacturing, consumer goods, and global social organizations.

We research and deploy technologies that power AI-native operations — both for our partners and for Distyl itself. Our work spans research into self-constructing systems, the development of the most reliable execution of AI systems, and products that transform mission-critical workflows. As a result, Distyl's technologies affect some of the world's largest operations — from hundreds of millions of consumer interactions to tens of millions of supply chain transactions and millions of patient journeys.

Distyl is backed by leading investors including Lightspeed Venture Partners, Khosla Ventures, Coatue, DST Global, and the board-members of 20+ F500s.

What We Are Looking For

At Distyl, Research Engineers build the bridge between frontier AI research and production systems that deliver real business value. This role is for engineers who are excited to investigate how AI systems should be designed, rapidly prototype new ideas, and turn promising concepts into reliable systems that work inside real customer environments.

Research Engineers operate at the intersection of applied research, systems engineering, and customer-facing deployment. They design and implement compound AI systems, run experiments to understand system behavior, build evaluation frameworks, and collaborate closely with AI Researchers, AI Engineers, and customer stakeholders. Their work is not limited to demos or isolated prototypes: they help turn new techniques into robust systems that can be measured, operated, and improved in production.

Key Responsibilities

Design and run post-training workflows that improve the behavior, reliability, and usefulness of AI systems

Develop datasets, preference signals, evaluation suites, reward models, fine-tuning workflows, and feedback loops for applied AI use cases

Investigate how different post-training techniques affect system behavior across enterprise workflows and production constraints

Build infrastructure for experimentation, model comparison, regression testing, and behavior analysis

Partner with AI Researchers to explore new post-training methods and with AI Engineers to apply successful techniques in deployed systems

Analyze model outputs, failure modes, human feedback, and production traces to identify opportunities for behavioral improvement

Create repeatable processes for adapting AI systems to customer domains while preserving robustness, transparency, and maintainability

Communicate clearly with internal teams and customer stakeholders about model behavior, evaluation results, limitations, and tradeoffs

Who You Are

Experience Improving Model Behavior: You have worked with fine-tuning, preference optimization, reinforcement learning, reward modeling, synthetic data, evals, or related post-training techniques

Strong Programming and Experimentation Skills: You can build training and evaluation pipelines, run controlled experiments, analyze results, and iterate quickly

Research-Oriented Builder: You care about understanding why behavior changes, not just whether a benchmark improves

AI Systems Mindset: You understand that model behavior is shaped by data, prompts, tools, retrieval, evaluators, and deployment context—not model weights alone

AI-Native Working Style: You use AI tools daily to accelerate coding, analysis, debugging, experimentation, and research exploration

Bias Towards Measurement: You make behavioral improvements concrete through evaluations, comparisons, regression tests, and production-relevant metrics

Comfort with Applied Constraints: You can balance research ambition with practical constraints around cost, latency, reliability, data availability, and customer requirements

Ownership Mentality: You take responsibility for whether post-training work improves real system outcomes, not just offline scores