Member of Technical Staff — Research

San Francisco / New York $120,000 - $250,000 + equity

AI has advanced by expanding what machines can represent.

Deep learning learned representations from raw data. Transformers gave machines access to the knowledge humanity compressed into language.

But descriptions are not experience.

Experts develop intuition by acting, observing consequences, and learning what matters. Models can consume descriptions of judgment, but they do not inherit the experience that produced it.

Collective Intuition is building the missing layer: infrastructure that makes human experience observable, learnable, and actionable by AI.

We deploy in the world's largest enterprises, where judgment is exercised repeatedly and tested against reality. Its record already exists in fragments: decisions, revisions, exceptions, approvals, failures, and outcomes. We make that experience learnable, allowing our models to inherit the judgment organizations have developed through consequence.

Language gave machines access to what humanity has said about the world. Collective Intuition gives them access to what happened when people acted within it.

The role

As our Founding Research Scientist, you'll define the research foundations behind this new layer of intelligence.

You'll work alongside a founding team with deep experience building and deploying production foundation models at Fortune 500s from scratch. This is a hands-on research role: you'll develop new methods, build working systems, make foundational technical decisions, and help establish our research culture.

The central challenge is learning stable, useful, and calibrated representations of experience from incomplete, noisy, and longitudinal observations. State is distributed across systems, actions are interdependent, outcomes arrive late, and causal attribution is difficult.

You'll define how behavior should be represented, how useful structure can be inferred from incomplete observations, and how uncertainty should be measured. You'll then take those ideas from research into systems deployed within real enterprise environments.

What you'll do

• Build models that transform noisy observational data into structured representations.
• Develop methods for learning organizational behavior directly from real-world activity.
• Model how organizational state and behavior evolve over time.
• Design uncertainty estimation and calibrated abstention systems.
• Create evaluation methodologies for problems where labels are sparse, ground truth is incomplete, and established benchmarks do not exist.
• Develop methods for evaluating the reliability under real-world conditions of learned representations.
• Take research from papers and prototypes into deployed systems while maintaining scientific rigor under real-world constraints.
• Work directly with the founders to shape our research direction, system architecture, and long-term technical roadmap.

What we're looking for

• A PhD or equivalent research depth in machine learning, statistics, computer science, or a related field.
• Research contributions in one or more of graph machine learning, knowledge representation, temporal or event modeling, process reconstruction, or uncertainty quantification.
• Strong statistical foundations, including calibration, uncertainty quantification, identifiability, and model evaluation.
• Experience designing evaluations for problems without established benchmarks or complete ground truth.
• Experience building learning systems where supervision is weak, indirect, or derived from structure.
• Strong engineering ability: you can build your own models, experiments, evaluation harnesses, data pipelines, and research infrastructure.
• Experience taking research ideas beyond prototypes and applying them within production or production-oriented systems.
• Comfort defining new problems and working with substantial autonomy where no established blueprint exists.

Useful, but not required

• Strong PyTorch experience; JAX experience is a plus.
• Experience working with noisy, weakly supervised, partially observed, or longitudinal datasets.

The next frontier in AI is the representation of experience itself.

Real work is a difficult learning environment: state is distributed, actions are often implicit, and outcomes arrive late. Making experience learnable requires solving hard problems across representation learning, graphs, temporal reasoning, uncertainty, and long-horizon credit assignment.

That is what we are building.

You'll work directly with the world's largest enterprises, taking foundational research into systems whose utility is immediate and measurable. Ideas will not end at papers or benchmarks: you'll see them change how consequential work is understood, governed, and improved at scale.

There is no established blueprint for this. You'll have meaningful influence over the research direction, the product, and the company built around it.

We are not building another interface to existing models. We are building the substrate from which the next generation of intelligent systems can learn.

#J-18808-Ljbffr Jobleads-US