Who is behind Wet Ink
Who is building it
Michael Silva builds Wet Ink. He previously built the governed agent platform that proved this kernel's enforcement pattern — a system where AI-produced work moves through database-enforced gates, evidence ledgers, and human approval before anything counts as done. Wet Ink rebuilds that discipline clean for the highest-stakes document judgments in home lending. The name comes from banking's most human artifact — the wet signature, the authoritative original. Every AI action gets its wet-ink moment: a distinct, accountable approval, in a record built to outlast the meeting where it happened.
What is claimed and not claimed
- The sealing architecture is real: an append-only, tamper-evident record with schema-enforced maker/checker separation, and
85passing proofs behind it. - The architecture's design basis is full-population reperformance — every file, each with its own evidence pack.
- We designed the record for recomputation by an outside verifier — trust the math, not the screen.
- No live loan review. The AI judgment layer does not yet exist; no loan file is under review today. See what runs today.
- No customers, no pilots, no deployments — and therefore no customer results, anywhere on this site.
- No accuracy percentages. None exists to quote, so none appears.
- No certifications and no auditor's conclusions — no "compliant," no "examiner-approved." Coverage evidence only, stated with its denominator.
- Not "tamper-proof," and not a blockchain. The record is tamper-evident — a per-tenant hash chain in Postgres — and the chain catches alterations rather than preventing them.
Press
For press and analyst inquiries, write directly — a named mailbox, not a form.
For the regulatory context Wet Ink answers, start withthe LL-2026-04 explainer — what the framework obligates seller/servicers to do, and what an answer built to be checked looks like.