Mihu AICustom dataset design
Datasets built to your specification: task design, schema, difficulty targets, rubrics and acceptance criteria.
Custom data for AI training & evaluation
Custom datasets, built by a network of domain experts.
Mihu AI designs and delivers custom training and evaluation data for AI research teams. We crowdsource each project from a vetted network of university students and graduates in the relevant field, with expert review on every item.
Services
What we deliver
Expert crowdsourcing
Projects staffed from our network of domain experts, matched by field and level of study and scaled to your volume.
Problems & worked solutions
Original, hard problems with step-by-step reasoning, written at university and graduate level.
Rubrics & grading
Task-specific rubrics and rubric-based assessment of model outputs.
Preference & evaluation data
Pairwise comparisons, rankings and written critiques of model responses.
Dataset review & correction
Expert audit of existing datasets: verifying answers, fixing errors and flagging ambiguous items.
Domains
- Mathematics
- Physics
- Chemistry
- Biology
- Computer science
- Engineering
- Economics & finance
- Humanities & social sciences
Approach
How we work
-
Specification
We agree the task design, rubrics and acceptance criteria with you before any work begins.
-
Expert sourcing
We crowdsource vetted undergraduate, master's and PhD-level contributors in the exact field from our expert network.
-
Authoring & review
Every item is written by one specialist and independently reviewed by another. A senior reviewer settles disagreements.
-
Delivery
Data is delivered in your format, with a record of who authored and reviewed each item.
Our expert network
University undergraduates, master's and PhD students and recent graduates across disciplines. Before joining a project, each contributor:
- has their enrolment or degree verified
- passes a field-specific screening task
- signs a confidentiality and IP assignment agreement
Trust
Security & compliance
Confidentiality
All engagements run under written agreements. Project materials are available only to assigned contributors, and we do not disclose client names or project details.
Data protection
We comply with Singapore's Personal Data Protection Act 2012 and protect personal data transferred overseas to a comparable standard. See our privacy notice.
Sanctions & export controls
We comply with applicable Singapore, UN and US sanctions and export control laws, and do not engage contributors located in comprehensively sanctioned jurisdictions.
Original work
Contributors may not use AI tools to produce deliverables unless a project explicitly allows it. Our review process checks for this.
Clear IP rights
Every contributor assigns intellectual property in their work to us, so clients receive clean rights to what we deliver.
Registered entity
Mihu AI Pte. Ltd. is incorporated in Singapore and registered with ACRA under UEN 202644374E.
Contact
Discuss a project
Tell us about the data you need. We will reply by email.
- Company
- Mihu AI Pte. Ltd.
- UEN
- 202644374E
- Registered office
- 152 Beach Road, #23-02
Gateway East, Singapore 189721