About
Kyo Harada — Founder of Visage AI / Freelance Data Scientist.
- CoreML + Vision for on-device skin analytics (iOS)
- Finance-grade data pipelines (EDINET XBRL, MySQL, dashboards)
- Bilingual delivery (JP/EN)
Research and open-source work
analysis-integrity-kit
Executable checks for analysis defects that can run to completion while returning plausible values, plus an operational specification for governed data-science work.
- Concept DOI: 10.5281/zenodo.21714101
- Version DOI (v0.1.0): 10.5281/zenodo.21714102
- GitHub repository
Preprint: other-agent models in LLM systems
Beyond Social Performance: Evidence Modes and Validation Obligations for Other-Agent Models in LLM Systems is published as a version 1.0 preprint on Zenodo. It defines evidence modes and the validation obligations attached to claims about other-agent models in LLM systems. The work is citable by DOI; it is not presented as peer reviewed.
Related manuscript in preparation
Governed LLM-Assisted Analysis: A Failure-Mode Taxonomy and an Operational Boundary Specification is in preparation. The current status and source materials are public; no completed manuscript or peer-review claim is made here.
EBM-2025 replication audit
The corrected EBM-2025 v0.2 report withdraws five claims from the September 2025 edition after a code audit. The original remains archived.
Contact
Work with me → Contact and engagement options