Distyl AI is looking to solve complex, high-stakes challenges at scale for Global Fortune 1000 companies by pioneering AI-native systems of work, requiring creative researchers to redefine how software is used.
Requirements
Deep Understanding of Post-training Techniques: Familiarity with supervised fine-tuning, preference optimization (RLHF/DPO), LoRA/PEFT, and instruction-tuning pipelines.
Experience Adapting Frontier Models: You’ve tuned or adapted LLMs/SLMs to specialized domains or behaviors through data curation, reward modeling, or continual pretraining.
Experience Building with Models, Not Just Building Models
Proven Track Record of Research Results
Uses AI Every Day
Strong Programming and Data Analysis Skills
Biases Towards Showing vs Telling
Responsibilities
Researchers develop and evaluate techniques such as supervised fine-tuning, preference optimization (DPO, RLHF, RLAIF), and continual adaptation to align models with Distyl’s enterprise systems.
The goal is to bridge raw model capability with trustworthy, contextually aligned system behavior.
Researchers in Post-Training investigate new methods for aligning large models with human and system-level objectives.
They explore trade-offs between generalization and specialization, data efficiency and robustness, capability and controllability.
Their work informs how Distyl leverages foundation models safely, effectively, and at scale across industries.
We develop intelligent systems using models rather than training or fine-tuning them.
Ideal candidates have expertise in compound AI systems, agentic collaboration, and associated techniques (ensembling, ReAct, graph-of-thoughts, etc.).
Other
This requires creative researchers who don’t just want to drive incremental improvements on benchmarks or optimize an existing process but instead are looking to creatively redefine how software is used.
Our researchers come from many academic backgrounds but have strong research track records, operate in an AI-native way, and would be bored staying on the rails of a traditional research org.
While you might not consider yourself a software engineer you need to be able to build prototypes of your ideas and then perform the experiments to prove the effectiveness to a F500 Head of AI.
Our customers want to see the power of AI today vs discuss the most elegant idea that will take 5 years to realize.
Distyl is a hybrid working environment and requires in office collaboration 3 days a week.
