AI you can check.
AI output is cheap. Trust is scarce. Mind Theory AI builds products for workflows where a wrong answer has consequences, so each one separates evidence from inference and shows its sources.
Three products, one rule: show the evidence
Part of the Ghostgraph.ai suite. Visitor screening for schools, hospitals and venues; investigation-grade vetting for PIs, corporate security and SIU teams.
Pricing published: Receipt Verified 2026-10-09Live screening is priced from $1,500 to $6,500 per site per year.PeriodLive pricing page, 2026-10-09MethodPublished price list.Sourceghostgraph.ai pricing page per site per year and Receipt Verified 2026-10-09Investigations are priced from $4,999 to $29,999 per seat per year.PeriodLive pricing page, 2026-10-09MethodPublished price list.Sourceghostgraph.ai pricing page per seat per year.
Plain-English California law for people who can't afford a lawyer for every question.
Receipt Verified 2026-10-09Search 126,650+ California statutes across all 29 codes, no sign-up required.PeriodLive site, 2026-10-09MethodThe live site's own description of its corpus.Sourcelawwiki.aiCaveatKnown data defects: at least 277 sections are truncated at 15,000 characters and some 2026 text is stale. statutes searchable, with quote-level citation checks.
Private, on-device translation for families and teams who switch between Thai and English mid-sentence.
English error rate on code-switched audio Receipt Verified 2026-10-09Language arbitration cut English word error rate on code-switched audio from 0.988 to 0.174, with 0 false switches.Period2026-09-24MethodFLEURS, 100 English + 100 Thai clips; paired bootstrap 95% CI on the change: -0.88 to -0.74.SourceEvaluation output (JSON)CaveatPublic benchmark clips, not real calls..
Founder
- Go-to-market and analytics leadership inside growth-stage software businesses and product lines.
- Machine learning in production: Receipt Verified 2026-10-09Trained 70+ Python machine learning models for churn risk.PeriodJan 2025 to Jan 2026MethodCount of production scoring models.SourceMaster resume (Sep 2026) models trained to get ahead of churn.
- Builds and ships the product myself.
How we build
- Receipt Verified 2026-10-09Shipped 16 merged pull requests and 60 commits on main in 11 days.Period2026-09-29 to 2026-10-09MethodMerged PRs and commits on the main branch.Sourcegit history and GitHub merged to production in 11 days, solo, with CI that scans all of git history for personal data.
- An AI-native build process: Claude Code does the typing; I write the specs, the acceptance gates and the verification.
- Evaluation discipline as a moat: negative controls, pre-registered rules and confidence intervals on every model choice.
Why verification matters
Screening, legal research and transcription fail the same way: a confident guess presented as fact. Each product makes that failure visible.
Inferred links are flagged and show the entity they came through. A screen that has never returned a positive is reported as unverified, not clean.
Every quoted line is checked against the statute text before it is shown.
An independent model must hear speech before the best transcriber is allowed to write anything.
Stage and the ask
Ghostline and LawWiki are live in production, and Ghostline's pricing is published. Nothing on this site claims customers, pilots or revenue.