SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Shuttlebase is building an embedding model for human behavior that predicts how people interact with websites and what optimizations will work before testing. The company processes tens of millions of user actions daily from e-commerce sites and aims to help any company optimize their digital experience for different audiences.
As the first full-time Applied Scientist focused on Behavior Modeling, you will own the development of Shuttlebase's core behavior model. Your responsibilities include:
**Core Work:**
- Build the first model that transforms raw session events into behavioral embeddings, likely starting with a transformer pre-trained on event sequences
- Develop methods to identify buying intent during sessions and segment visitors into audiences for personalized page experiences
- Deploy an initial version to live stores and iterate based on real-world performance
- Stay current with published research in sequence modeling and embeddings, evaluate new approaches against your data, and integrate promising techniques
- Publish research findings when appropriate
**Collaboration:**
You'll report directly to the CTO and work closely with engineers who build on your models and client teams who communicate results to e-commerce partners. You'll be responsible for explaining model decisions and insights to non-technical stakeholders.
**Technical Environment:**
Python-based development with event data stored in ClickHouse and services deployed on AWS. You'll write SQL to access raw data and work with noisy, real-world event logs.
**What Success Looks Like:**
You thrive on reading research papers and immediately testing ideas on real data. You prefer adapting proven methods that work in weeks over inventing novel approaches that take years. You want ownership of a significant, open-ended problem with direct access to founders and engineers. You're comfortable with messy data and can translate complex ideas for non-technical audiences.
**Requirements:**
- PhD or equivalent research experience in a quantitative field (physics, math, computational biology, neuroscience, or similar)
- Hands-on experience training models on messy, real-world sequence or time-series data with demonstrated validation of results
- Deep knowledge of recent sequence learning work: transformers, self-supervised pre-training, embeddings—enough to choose between approaches and justify decisions
- Solid Python programming skills and ability to write SQL to extract your own data
- Demonstrated ability to communicate complex ideas to audiences outside your field (talks, papers, or similar)
- Bonus: background in recommendation systems, user behavior analysis, large-scale event datasets, or e-commerce
