Learning data, written by people who teach.
A tutor that gives away the answer, or explains it wrongly, does harm at scale. Education models need data from people who know both the subject and how students learn.
Education, with proof.
Edtech companies and learning platforms build models that tutor, explain, generate practice questions, give feedback on writing and grade open responses. The failures are pedagogical as well as factual: skipping the step a student is stuck on, praising a flawed argument, setting a question at the wrong level. HUMXN commissions credential-checked teachers, subject experts and linguists to write worked solutions, tutoring dialogues and feedback, and to grade work against your rubric. You set the subjects, levels and teaching approach. We scope it and deliver a pilot batch first.
Grading is where expert data pays for itself. Automated feedback must agree with a skilled teacher often enough to be trusted, and that only happens if the training and evaluation data were graded carefully. Our educators grade to your rubric, write the feedback a student would actually see, and record their reasoning. Labels are marked by source, and contested grades go to a second expert. Student-style responses can be written by educators to reflect common misconceptions, so a dataset does not depend on real student work.
Where documents or tables contain personal data, they are scanned and masked before delivery, and personal metadata never reaches buyers. Licensed textbooks, problem sets and educational media from the catalogue carry their AI-training permission and licence terms. Each item has an Ed25519-signed provenance record and appears on a signed receipt by ID and hash. That supports the privacy and procurement reviews schools and districts run, including under frameworks such as FERPA and GDPR, without replacing your own assessment.
Built to your specification.
Every engagement starts from a written spec and a pilot batch. These are the most common requests we source for education.
Worked solutions
Step-by-step solutions in mathematics, science and other subjects, written at the level you target and reviewed by a second expert.
Tutoring dialogues
Educator-written conversations that guide rather than tell, with hints, checks for understanding and recovery from misconceptions.
Rubric-graded responses
Essays and open answers graded to your rubric by experienced teachers, with the written feedback a student would receive.
Misconception-based items
Student-style answers written by educators to reflect common errors, for training feedback models without real student data.
Curriculum-aligned question banks
Original practice and assessment items written to your standards and levels, exclusively licensed if you need them.
Multilingual learning content
Explanations and items written or translated by subject experts and linguists for learners across languages.
01Answers instead of teaching
Models trained on answer keys hand out answers. Educator-written dialogues teach the model to scaffold.
02Inconsistent grading
Feedback models inherit their graders' inconsistency. A shared rubric and second-expert review make grades dependable.
03Student privacy
Educator-written student-style work avoids real student data, and documents are scanned and masked for personal data.
Education: asked often.
Who writes and grades education data?
Credential-checked teachers and subject experts, matched to the subject and level in your spec. A secondary mathematics rubric is graded by people who teach secondary mathematics. Each label records whether it came from a creator, an AI suggestion or expert review, and contested grades go to a second expert.
Do you use real student work?
Most education datasets we scope use student-style responses written by educators to reflect common misconceptions, which avoids real student data. Where documents or tables contain personal data, they are scanned and masked before delivery, and personal metadata never reaches buyers. Your privacy team should assess any dataset against obligations such as FERPA or GDPR.
Can items align with our curriculum or standards?
Yes. The dataset spec names the standards, levels and teaching approach, and experts write to it. The pilot batch lets your curriculum team check alignment and tone before volume. Commissioned items can be delivered under an exclusive licence so they stay unique to your product.
How is education data priced?
By subject, level, item type, review depth and licence. A multi-turn tutoring dialogue costs more than a single practice question, and second-expert grading adds cost. Volume and standing orders reduce the unit price. We quote once the spec and rubric are agreed.
What formats do you deliver?
Dialogues, solutions and graded responses as JSON Lines or Parquet, question banks as CSV or JSON Lines, and media in WebDataset shards, with Croissant 1.0 metadata. Every delivery includes a signed receipt listing each item by ID and hash, and approved buyers can use the API.
Tell us what your model needs to learn.
Send a brief in five minutes. A data lead replies within one business day with questions and a first sourcing plan.