Loan-Pack Checker
Checks every loan pack before disbursal — key fact statement, sanction letter, agreement and KYC set — and shows each mismatch with its page.
Every pack checked before disbursal, held packs first, each with what's wrong in a few words
The real screen, on synthetic dataThe KFS rate and the agreement disagree: both pages side by side, and the pack is held
The real screen, on synthetic dataThe annual percentage rate recomputed in code from the pack's own schedule and fees, with its working
The real screen, on synthetic dataA page missing from the scan: the pack is held and nothing is filled in from another document
The real screen, on synthetic dataThe whole demo as a video, with captions and chapters
Demo built on synthetic data.
1 of 5: The pack queue
The problem
The key fact statement gives one annual rate, the agreement implies another, and the fee schedule doesn't add up to either. Credit operations check this by eye, pack by pack, and a mismatch found after disbursal becomes a complaint.
What it does
It reads every document in the pack into fields, each with the page it came from.
It checks the documents against each other and recomputes figures such as the annual percentage rate from the pack's own schedule.
It clears clean packs, sends mismatches to operations, and holds a pack when something critical is off.
It produces a daily exceptions report.
Try to break it
On the call we'll trigger these, live:
- A page is missing from the scan. The pack is held.
- Two versions of the agreement are uploaded. The conflict is shown before anything clears.
- The model server stops. Packs wait; none are cleared automatically.
Where it runs
Your data centre or an India-hosted cloud account in your name.
Who else does something like this
Finnulate sells AI-based compliance tooling for RBI's lending directions.
The Checker works on each pack before disbursal and runs on your own infrastructure.
The rules it respects
Built around the key fact statement requirements in RBI's lending rules. A pack that doesn't match its own KFS is held.
What our tests measured
How we testThese figures come from test runs on Google's Gemini (gemini-3.5-flash-lite), a hosted model, on synthetic data. We'll add runs on a local model.
- Run on 14 October 2026
- Model: gemini:gemini-3.5-flash-lite
- 200 examples
For your engineers
- Engines
- Extractor, Gatekeeper, Proof Console
- Sealed mode
- Test passed: with sealed mode on, it refuses every outbound connection except to the model you've approved.
Data flow