roger3000 · demo — the intelligence layer between your data and your CRM · meridian advisory (fictional)
roger3000 reads what you already own — CRM, documents, database — runs investor intelligence over it, and writes the enriched result back where you work. And it tells you when it's guessing: every claim is audited against a source before you see it. Six moments from the machine.
Your CRM, your documents, your database — read in place, not an integration project. The engine reads them into one typed graph: firms, funds, partners, deals, and how they connect.
External market signal — fund closes, deals, team moves — arrives already joined to your entities. Not a feed of headlines to triage; a fact attached to a firm you know, scored against a mandate you wrote.
Every deck becomes a saved artefact — its own workspace, kept per pitch. The engine reads it, matches it against the firms it knows, and returns a ranked report: the investors that fit, each with the signal that put them there.
Not a list of names — a report you can stand behind, because every claim in it is checked before you see it.
Every sentence the engine writes is checked against the sources it holds. Verified claims carry their evidence. Unverifiable ones are kept — and flagged as such, so you know exactly where the ground is. Fabricated ones never reach you.
And the auditor itself is measured. On a fixed labeled set it currently calls nothing grounded that isn't — 0.0% fabricated false-positives — and catches every planted fabrication. Most AI asks for your trust; this one shows you its score.
Ask in plain language. The brain answers from your own documents, and every claim cites the source it came from. If the documents don't hold the answer, it says so and abstains — it never fills the gap with a guess.
So when your premise isn't in the sources, it won't play along — it shows you what the documents actually say. The same grounding standard the claim auditor holds your reports to in 03.
The engine writes the enriched result — the fit, the warm path, the tracked signals — straight back into your CRM record, so the tool your team already lives in starts surfacing deals it couldn't see. And it reads their moves back: a stage change, a new contact — the layer stays in step.
Every write is previewed, dedupe-safe, and yours to approve — it never touches your pipeline on its own.
The outcome happens in your CRM — a pass, a meeting, a term sheet. The layer reads it back and turns it into a labeled example: this startup, this firm, this result.
The matcher trains on those labels. That's the part no generic model can copy — it isn't in anyone's training data. It lives in the layer, on your outcomes, and it compounds.