CaliKo
We build the architecture beneath AI that has to get it right — grounded, cited, defensible.
Principal
CaliKo is run by Min Chang. Twenty years of practice building systems that have to be right — equipment twins at GE, worker metrics at Workday, account-based selling at Salesforce, company intelligence at BETA, professional personas at pxtxt. CaliKo applies that craft to AI.
What we take on
We take on three kinds of problems — for teams building AI where being right matters more than sounding right.
-
Knowledge capture where AI can't afford to hallucinate.
Turning a domain expert's fluent knowledge into grounded, cited AI input is labor-intensive and easy to get wrong. We build the interview, extraction, and curation pipeline that does it without flattening the expert.
-
Cross-validation for AI that makes claims about real entities.
Unverified claims about people, companies, or institutions compound into liability fast. We build multi-source verification pipelines that keep those claims defensible under audit.
-
Persona-strict representation that says "I don't know."
When a generative model represents a specific person or institution, fluent-but-wrong is worse than a bounded "I don't know." We build representations that speak only from curated input — and refuse to fill the gaps.
What we've demonstrated
BETA and pxtxt are demonstrations of the craft, not client case studies. We built them on our own problems first.
-
BETA
Company intelligence with multi-source citation — every claim holds up to audit. Demonstrates the cross-validation pattern applied to organizations.
-
Pixels & Text
AI-guided interview, extraction, and compilation for professional personas. Demonstrates the interview-to-canonical pipeline applied to people.
How to engage
A conversation starts with a real problem. Tell us what you're building and where you're stuck.