Three leaders from across Two Sigma appeared on podcasts in recent months to discuss how the company approaches data, technology, and management. Their conversations reveal core themes that define how we work: building powerful platforms that enable continuous innovation through cutting-edge technology and a deeply scientific mindset.
Across these discussions, several common principles emerged:
- Platform-centric approach: We’re fundamentally a “platform company” where foundational infrastructure enables rapid experimentation and innovation
- Balancing speed with rigor: Moving fast to capture opportunities while maintaining the scientific and quantitative rigor required in investment management
- Technology as enabler: Viewing emerging technologies (e.g., LLMs and AI agents) as tools that amplify human creativity rather than replace it
- Continuous adaptation: Our competitive advantage comes from constantly evolving our platforms and approaches as new technologies emerge
The conversations covered diverse ground:
Effie Baram, Foundational Data Engineering Lead (Data Engineering Podcast) explained how our foundational data team moved from shipping only production-ready datasets to offering both raw and curated data, letting researchers start work immediately while building out full historical depth and quality in parallel.
The team standardized on modern warehouses like BigQuery, treats data transformations as code with full CI/CD pipelines, and focuses on “ice cube” patterns: well-defined, reusable data contracts that meet 90% of use cases rather than building custom solutions (“snowflakes”) for every request.
Ben Wellington, Deputy Head of Feature Forecasting (TWIML AI Podcast) discussed how LLMs have the potential to collapse feature engineering tasks from months to minutes. For instance, a modeling team could test whether a CEO touching their nose during a TV interview predicts stock movements—and could do so as easily as they once counted word frequencies in a transcript.
He also emphasized the critical importance of controlling for temporal leakage by using open-source models trained on point-in-time data, ensuring models don’t “know” future information when making past predictions. Looking ahead, he explores how different LLMs could provide orthogonal signals, imagining a future of “combinatorial research” with potentially thousands of AI agents discovering new patterns.
Matt Greenwood, Chief Innovation Officer (Dev Interrupted) reflected on 20+ years of helping to build Two Sigma’s platform, from yanking hard drives out of vendor machines to managing millions of lines of code today, describing innovation as “stacked S-curves” where each platform reveals entirely new problems you couldn’t see before.
He advocates an “epsilon and omega” approach—maintaining long-term vision while taking small meaningful steps that generate learning—and categorizes AI usage into four roles (advisory, oracle, operational, and agentic), emphasizing that automation should free humans to express ideas rather than replacing them altogether. (See here for a deeper dive into this episode.)
Next Steps
To learn more, check out Effie, Ben, and Matt’s podcast appearances. And if you’re curious about building platforms that push the boundaries of quantitative finance, visit Two Sigma’s Careers page.
