AI & ML interests
Education, Tutoring
Recent Activity
National Tutoring Observatory
The National Tutoring Observatory (NTO) at Cornell University studies how tutoring works at scale and shares what it learns. This organization publishes Million Tutor Moves (MTM): a growing, openly licensed corpus of de-identified K-12 tutoring sessions from partner providers, released in numbered data releases over three years.
🌐 nationaltutoringobservatory.org · 💻 GitHub
Releases
| Release | What it is | Status |
|---|---|---|
mtm-000-sample |
A small sample: documentation plus four human-reviewed math sessions (group video, one-on-one voice and text chat). A taster for the series. | available (2026-11-16) |
mtm-001-… |
Text corpora from partner providers. | planned |
mtm-002-… |
Audio-video sessions processed by NTO's multimodal and de-identification pipeline. | planned |
Each release has its own page with its schema, its de-identification method, known issues, a changelog and a citation. The list above is the only place releases are announced.
How access works
All MTM data are licensed CC BY 4.0. To download any release you accept the Million Tutor Moves Data Use Terms once, on any release page, where they are shown in full; the acceptance covers every current and future release. Approval is automatic. In short, the terms ask you not to try to identify anyone or recover what was removed, not to use the data to harm or make decisions about individuals, to report any identifier that looks genuine, and to link to a release rather than re-host it.
Found something that looks like a real name or place? Email sandpiper_admin@cornell.edu. Don't post it.
De-identification
Every release says on its page how it was de-identified. Release 0 was de-identified by hand: names, places and other personal details were replaced with realistic stand-ins (surrogates), so the text still reads naturally, and every row was then reviewed. Later releases go through NTO's automated pipeline and an independent re-identification review. De-identification is never perfect; the terms and each page explain what that means for you.
Also here
MathEd-PII: a benchmark for detecting personal information in math-education text, with its paper, Utility-Preserving De-Identification for Math Tutoring (arXiv:2602.16571).- De-identification models used in NTO's pipeline are private research checkpoints and are not released.
Contact
Million Tutor Moves and this organization: Doug Pietrzak (technical director), Kirk Vanacore (research), René Kizilcec (director). General: sandpiper_admin@cornell.edu.
Acknowledgments
This work is supported by the National Science Foundation (Grant No. 2321499), the Gates Foundation, and the Chan Zuckerberg Initiative. Any opinions, findings, and conclusions are those of the authors and do not necessarily reflect the views of the funders.