Research Scientist, Modeling

Tenera, Inc; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Benefits offered by this job

Equity

Relocation assistance

Job summary

Tenera, Inc is seeking a Research Scientist in Modeling to build population models from user data and deploy them into customer deployments. You will design and train models from traditional ML to transformer architectures, using dedicated compute resources to iterate based on deployment feedback.

You will work closely with Forward Deployment to turn modeling ideas into validated solutions, balancing practicality with rigor and communicating results clearly to customers.

Qualifications

• Applied ML work with measurable results and explanations of choices.
• Hands-on pretraining of models from scratch including data prep and objectives.
• Experience designing and training transformer models.

Responsibilities

• Translate customer needs into population models and segments.
• Build and train models end-to-end, including custom transformers.
• Develop traditional ML models with strong baselines and evaluations.
• Create training datasets and sampling strategies from diverse data.
• Run controlled experiments and ablations with the evaluations team.
• Collaborate to deploy models into customer deployments and reusable platform capabilities.

Skills

Applied machine learning

Pretraining models

Transformer models

Python

Experimentation

Model evaluation

Collaboration

Tools

PyTorch

scikit-learn

Job description

Tenera is a human behavior simulation lab. We build simulated users that help teams predict how people will react to a product change before it ships.

As a Research Scientist in Modeling, you will work closely with Forward Deployment to build population models for real customer problems. This is an applied role: you will turn data about users and their behavior into models that can support customer decisions, then improve those models using evidence from deployments.

You will build and train models yourself, from traditional machine learning methods to custom transformers and smaller models pretrained from scratch. We value creativity and independent judgment: you should be able to explain why a method fits the problem, test that reasoning, and change your approach when the evidence points elsewhere.

We value practical models that work reliably on real customer problems. Choose the simplest approach that delivers the result, and add complexity only when the evidence shows it improves outcomes.

You will have dedicated compute resources to train models, run experiments, and iterate on your research.

02 What You’ll Do

• Work closely with Forward Deployment MTS to understand each customer's population, available data, and product decisions. Translate those needs into population models that represent differences across users and segments.
• Build and train models end to end. Design custom transformer architectures, pretrain smaller models from scratch, and fine-tune existing models when they fit the problem.
• Develop traditional machine learning models, including regression, tree-based methods, clustering, and probabilistic models. Establish strong baselines and compare approaches on predictive quality, calibration, interpretability, latency, and cost.
• Develop training datasets and sampling strategies from customer data, observed outcomes, and synthetic data. Account for population coverage, data quality, and separation between training and evaluation sets.
• Run controlled experiments and ablations with the evaluations team. Test generalization across populations and product contexts, investigate failure cases, and use the evidence to choose what to improve next.
• Partner with Forward Deployment and platform MTS to put models into customer deployments and turn useful approaches into reusable platform capabilities. Own the path from a modeling idea to a working, validated solution.

03 What We’re Looking For

• A track record of applied machine learning work that led to working models and measurable results. Show us the problem, the modeling choices you made, and the evidence that your approach worked.
• Hands-on experience pretraining smaller models from scratch is required, including preparing training data, choosing objectives, running training, and diagnosing optimization problems. Experience fine-tuning existing models is also expected.
• Experience designing, implementing, and training your own transformer models is required. You understand the architecture and training process well enough to modify them and explain the tradeoffs.
• Practical experience building traditional machine learning models is required. You can work confidently with regression, tree-based methods, clustering, or probabilistic approaches and choose suitable methods for the data and task.
• Independent thinking and creativity. You develop your own hypotheses and modeling approaches, question popular assumptions, and choose methods because they solve the problem rather than because they are the latest LLM trend.
• Strong Python skills and experience with deep learning and traditional ML tools such as PyTorch and scikit-learn. You can debug training runs, build reproducible experiments, and contribute production-quality code.
• Sound experimental judgment. You can prevent data leakage, reason about sampling bias and uncertainty, and test whether improvements generalize to the populations a customer needs to understand.
• The ability to collaborate closely with Forward Deployment and customers. You can turn practical requirements into modeling decisions, explain limitations clearly, and iterate quickly as new evidence arrives.
• Humble. You are quick to learn from customers, teammates, and evidence instead of defending your first answer.
• Low ego. You care more about the best idea winning than being personally right.

Tenera is an in-person company, working 5 days a week in our SF office. Open to relocation. Meaningful equity for the right person.