AI for what should
be automated.
People for what shouldn't.

No, and we’ve spent eighteen months proving why. We tried single-pass LLM inference early in development; it produced large deviations from a senior officer benchmark — the difference between a sound decision and a marginal one. So we built something different.

Calliper is a multi-stage pipeline, not a chatbot. Mathematical work — rate calculations, debt service ratios, covenant math — runs on deterministic code, because correctness isn’t negotiable. Document parsing, risk synthesis, and analytical reasoning run on AI models we host ourselves on Canadian hardware. A scoring framework we built specifically for community lending files replaces the generic scorecards that don’t survive contact with thin-file borrowers.

Our methodology validates within five basis points of a senior officer benchmark. And every file is then read end-to-end by a senior underwriter who stakes their professional judgment on the recommendation.

ChatGPT writes essays. Calliper writes credit memos.

No. Calliper prepares the analysis. Your loan committee and board make the call. Every time.

Calliper runs on our own GPU infrastructure in Cornwall, Ontario — not on commercial cloud AI services. Personally identifiable borrower information never leaves our infrastructure. A sanitisation layer strips identifying details before any external service is called, and our reasoning models run locally on Canadian hardware.

We’re built for OCAP, PIPEDA, Quebec’s Law 25, and OSFI E-23. Data residency is Canadian. SOC 2 Type 2 audit is planned.

Every file is signed by a senior underwriter who stakes their professional judgment on the recommendation. You’re not getting a model’s output — you’re getting our underwriting opinion, with a full audit trail back to source documents. If something doesn’t sit right with you, we walk you through the reasoning. If we got something wrong, we own it.

No, and we’d push back if you tried to use it that way. Your team does work Calliper can’t: site visits, relationship calls, the judgment that comes from knowing a borrower for years. What Calliper takes off their plate is the synthesis grind — reading two hundred pages of source documents, building the data manifest, drafting the credit memo. They get those hours back for the work that actually requires them in the room.

Speed: five business days instead of three weeks, often as fast as forty-eight hours. Consistency: the same scorecard every time, calibrated to your policy. Price: fixed per file, no day rates. Capacity: we don’t go on vacation in July. And depth: because the platform does the heavy synthesis, our underwriters can go further into sector context, sensitivity testing, and covenant analysis than a contract reviewer working alone ever could in the same window. You get a level of analysis that simply isn’t economical any other way.

Annual configuration reviews are included. Ad-hoc policy updates between reviews are handled as part of onboarding support at no additional per-file cost.

Nothing to download, nothing to set up. You sign in with a one-click magic link and upload a complete file through a secure portal — no integrations with your existing systems required.

You buy file credits in advance — one at a time during the pilot, or in 5- and 10-packs. A credit is consumed when you submit a file for underwriting, regardless of how the file ultimately moves through the pipeline.

If we find gaps during intake, we send the gap list within 48 hours and you resubmit when ready. Resubmission of the same file within 90 days does not consume an additional credit.

Before submitting, your team can use our intake checker to confirm a file has everything we’ll need — we walk you through it during onboarding.

Calliper is live at ACCFutures today, with pilots underway at additional CFDCs across Ontario. References are available on request once we have scoped a pilot together.