A chat + canvas AI tool that drafts credit memos starting from a bank's own template — designed to work across loan origination systems, and to hand back a polished, bank-branded document at the end, not just an unfinished draft.
A credit memo justifies a lending decision — ratios, borrower history, a recommendation, in the format a credit committee expects. It's repetitive work: analysts reconstruct a familiar structure by hand for every deal. And banks run different origination systems and already have their own memo templates, so a tool that forced a generic format or worked in only one system would start a fight it didn't need to have.
The chat panel takes plain-language instructions and corrections. The canvas is the actual memo — read and edited section by section, not a wall of text taken on faith. Anything the agent isn't confident about is flagged in place, and that section can't move to sign-off until the analyst resolves it.
Upload a format or pick a saved one; the agent drafts into that structure.
Sits alongside whatever loan origination system a bank already runs.
Analyst approves first; a credit officer signs off after.
Once approved, the memo compiles into the bank's own branded format.
The core question was how to split conversation from document so the analyst always knows where to intervene. Rapid prototyping tested an all-chat version, an all-canvas version, and combinations between — landing on chat-directs / canvas-verifies. The same iteration surfaced a smaller decision: a single uniform flag felt too blunt once tested side by side, leading to distinguishing a routine "needs review" from a genuine data conflict.
Being LOS-agnostic — rather than a feature bolted onto one origination system — makes it useful to a much wider set of banks from day one, and keeps the credit team's workflow independent of whatever core system the institution runs.