Founding Research Scientist, Forecasting Foundation Models (FFM)

About Torchcast

Torchcast builds AI systems that forecast real-world outcomes.

Our platform combines forecasting-trained models, evidence retrieval, quantitative methods, and probabilistic reasoning to help organizations make better decisions under uncertainty. We work across financial markets, economics, supply chains, technology, and other complex domains.

Our models currently hold the #1, #2, and joint #3 positions on ForecastBench's dataset leaderboard.

We are now developing Forecasting Foundation Models (FFM): a new generation of models purpose-built to learn from diverse evidence, reason under uncertainty, and produce accurate probability forecasts.

The role

We are looking for a Founding Research Scientist to help build Torchcast's FFM research program.

You will work directly with the founders and engineering team to develop new machine-learning approaches for forecasting. You will own research projects from initial formulation through experimentation, evaluation, and deployment.

This is a hands-on role for someone who combines strong research judgment with the ability to build working systems. As an early member of the team, you will have significant influence over our research direction, technical standards, and future hiring.

What you will do

• Research and develop new methods for probabilistic forecasting.
• Build models that learn from text, time series, structured data, market signals, and outputs from other models.
• Develop training and evaluation datasets using timestamped evidence, historical forecasts, and resolved outcomes.
• Design rigorous experiments and establish strong language-model, quantitative, time-series, and ensemble baselines.
• Evaluate models for forecast accuracy, calibration, robustness, and generalization.
• Improve how forecasting systems combine multiple sources of related, conflicting, incomplete, or duplicated information.
• Investigate forecasting-specific training objectives, architectures, and representation-learning methods.
• Work closely with engineering to scale promising research and integrate it into production systems.
• Contribute to technical reports, research publications, benchmarks, and open-source releases where appropriate.
• Help define the long-term FFM research roadmap and support the growth of the research team.

What we are looking for

• A strong track record in machine-learning research or the development of technically novel ML systems.
• The ability to take an open-ended research problem from formulation through implementation and rigorous evaluation.
• Deep expertise in at least one relevant area, such as probabilistic machine learning, representation learning, multimodal learning, time-series modeling, foundation models, uncertainty estimation, or structured prediction.
• Strong programming skills and practical experience with PyTorch, JAX, or a comparable machine-learning framework.
• A strong understanding of experimental design, model evaluation, temporal validation, and data leakage.
• The ability to identify high-value experiments and distinguish meaningful improvements from benchmark noise.
• Comfort working independently in a fast-moving, early-stage environment.
• Clear written and verbal communication skills.
• Strong research judgment, intellectual honesty, and a willingness to change direction when evidence does not support an initial idea.

A PhD in machine learning, computer science, statistics, applied mathematics, or a related field is preferred but not required. We value research ability, originality, technical judgment, and execution more than credentials alone.

Nice to have

• Experience with probabilistic forecasting, calibration, or proper scoring rules.
• Experience with multimodal or multi-source learning.
• Experience with time-series or temporal foundation models.
• Experience training or adapting large-scale models.
• Familiarity with graph, set-based, mixture-of-experts, or retrieval-augmented architectures.
• Experience with reinforcement learning or sequential decision-making.
• Knowledge of financial, economic, geopolitical, scientific, or operational forecasting.
• Publications at leading machine-learning, statistics, or related research venues.
• Experience moving research systems into production.

You do not need to meet every qualification listed above.

Why join Torchcast

• Join as a founding research hire with meaningful ownership over the direction of FFM, from the earliest research questions to the systems we ultimately build.
• Work on an ambitious technical problem at the intersection of machine learning, probabilistic reasoning, temporal dynamics, and real-world decision-making.
• Develop research on live forecasting problems, where predictions are ultimately tested against what actually happens.
• Move successful ideas beyond papers and prototypes into products used to support real decisions.
• Work directly with the founders and a small, highly technical team where strong ideas can quickly shape company direction.
• Help define the future research team, technical roadmap, research standards, and culture at Torchcast.
• Play a central role in establishing forecasting as a distinct and important direction for AI research.

How to apply

Please send the following to [email protected]:

• Your CV or resume
• Links to relevant papers, code, technical writing, or research projects
• A short note explaining your interest in Torchcast and forecasting research

Please also include a response of no more than 300 words to the following question:

What is one important research question in AI forecasting that you would be interested in investigating, and how would you begin testing it?

We welcome candidates from unconventional backgrounds and encourage you to apply even if you do not meet every qualification listed above.

Apply via [email protected]