AI Research Scientist / Engineer – Generative World Models

Teton is building the foundational data layer for the point of care, using real-time, multimodal AI to generate a digital twin of the patient and care environment. Our proprietary computer-vision system enables staff to automate documentation, proactively manage resident acuity, and streamline workflows across entire facilities. Our platform delivers unprecedented access to real-time data insights, empowering healthcare providers with actionable intelligence to enhance decision-making and elevate care delivery.

We are looking for an AI Research Scientist / Engineer to accelerate the development of Cassandra, our Generative World Model for healthcare.

Cassandra is pretrained on sequences of behavioral tokens representing residents' activities, sleep, mobility, vital signs, and care interactions. By learning to predict what happens next, Cassandra learns patterns of human behavior, enabling applications such as predicting falls before they happen and simulating the potential impact of care interventions.

Much like large language models, Cassandra learns from large-scale sequential data and exhibits similar scaling laws. However, modeling human behavior introduces unique research challenges, including continuous-time modeling, long-context dependencies, and learning from heterogeneous behavioral signals.

Unlike most foundation models trained on publicly available data, Cassandra learns from a proprietary and continuously growing dataset of real-world human behavior. This creates an opportunity to explore fundamental research questions while building technology with direct real-world impact.

We are looking for someone with strong technical intuition and hands-on experience training and scaling large sequence models. You will have the opportunity to take ownership of fundamental research challenges and directly shape both the underlying foundation model and the applications built on top of it.

Read more about Cassandra: Introducing Cassandra

In this role, you will own and drive research and development across two main areas:

• Pre-training and scaling Cassandra. Improve the end-to-end pre-training pipeline, from behavioral data representation and tokenization to model architecture, training objectives, and scaling. Many techniques from the LLM research toolbox are directly applicable, but important challenges remain, particularly around continuous-time event modeling, irregularly sampled data, and long-term dependencies. You will design experiments, develop new approaches, and translate research into improvements in model capabilities.

• Developing and productionizing downstream applications. Our first application of Cassandra's learned representations is predicting the risk of a resident falling within the next five minutes. You will improve these predictive capabilities and explore new applications, from forecasting changes in resident behavior to simulating the effects of care interventions. The goal is to turn advances in foundation modeling into tangible improvements in personalized, proactive care.

Minimum Qualifications:

• Hands-on experience training large-scale sequence models, such as Transformers, state-space models, or LLMs.

• Deep understanding of neural network architectures, optimization, training dynamics, and scaling laws.

• Strong Python and PyTorch skills, with experience building efficient, reliable training and evaluation pipelines.

• Ability to independently design experiments, interpret results, and translate research ideas into working implementations.

• Solid mathematical foundations in machine learning, probability, and statistics.

Mindset:

• Research-driven and hands-on. You enjoy reading papers, challenging assumptions, and implementing ideas yourself rather than stopping at theoretical insights.

• Strong technical intuition. You can reason about why a model works, identify bottlenecks, and prioritize experiments that are most likely to move the needle.

• Ownership and autonomy. You are comfortable tackling open-ended problems, navigating uncertainty, and taking ideas from initial hypothesis to production.

• Pragmatic and ambitious. You care about advancing the state of the art, but ultimately measure success by whether your work enables capabilities that were previously impossible.

Time Zone/location:

• Based in Copenhagen, Denmark, or willing to relocate.

Preferred Qualifications:

• Research or engineering experience with LLM pre-training, including architecture design, data mixtures, training objectives, or scaling experiments.

• Experience with generative modeling of sequential, temporal, or event-based data.

• Familiarity with continuous-time models, point processes, time-to-event prediction, or survival modeling.

• Experience with long-context architectures, such as Mamba, recurrent models, or efficient attention mechanisms.

• Experience with distributed training, GPU optimization, and performance profiling.

• A track record of impactful ML research, demonstrated through publications, open-source contributions, or production systems.

• Competitive Salary and participation in our warrant program.

• Work with state-of-the-art technology in a pioneering field.

• Large in-house training cluster.

• In-office work place.

• A vibrant, learning-focused work environment.

We're a growing team of extremely hard-working and talented people. The learning curves are steep, and expanding your skill set is not just encouraged — it's expected. It's a hands-on environment where you'll be challenged, supported, and constantly learning.

We are looking for people who believe in our long-term vision and value ownership and entrepreneurship rather than just another 9-5 job. With us you will have an opportunity to truly make an impact on the world with the outcomes of your work. So, if you are looking for a ride and not just a job – jump on board 😊