Strata
Analytical AI
Strata reads decades of the world's largest studies on human skill and learning, uncovering what teams of researchers would take years to see. Where one generation gained ground. Where a nation quietly fell behind. Where a single policy decision changed everything that followed.
Coming soon.
What we're building
The OECD's Programme for the International Assessment of Adult Competencies (PIAAC) is one of the richest datasets in existence on how adults around the world read, reason, and solve problems — and one of the hardest to work with. Its survey weights, plausible values, and thousands of coded variables mean that answering even a simple question about it usually takes a trained researcher days of careful, manual work. Strata is an agentic research pipeline built to do that work at machine speed, without cutting the methodological corners a human expert wouldn't.
System overview
How it comes together
Understanding the codebook. Teaching the system to find the right variable in OECD's dense documentation on its own, using retrieval over the codebook rather than a hardcoded lookup table.
Letting the agent do the analysis. An agentic workflow that plans a statistical approach, then writes and runs the actual Python needed to carry it out, applying the survey weighting and plausible-value adjustments PIAAC requires.
Checking its own work. Every finding the system reports is cross-checked against the deterministic statistical output that produced it, so a claim never outruns the numbers behind it.
Beyond PIAAC. Once the pipeline is proven on one large-scale assessment, extending it to others — like PISA — is the natural next step.
The researcher behind it

Sulakshna Devi Umrit
Data Science Intern, Traveling Waves
Sulakshna is a quantitative researcher bridging advanced statistical methodology with educational sciences. Her Master's research at the University of Oslo focused on analyzing PIAAC data in Norway using multilevel and intersectional methods (including MAIHDA), and she brings that same rigor to Strata's verification pipeline — making sure that what an AI reports about this data is held to the same standard as a published finding. She's driven by a broader interest in using data to surface and address systemic educational inequality.
LinkedIn ↗