1
Upload
Apprentice
Reads annual reports, financials, earnings material and net-worth statements, or pulls them from your data provider.
You
Tag each file as borrower, promoter or guarantor.
Raw files in.
A credit memo
you understand.
Apprentice builds the CMA, reasons over the numbers the way a seasoned RM would, and cites the page behind every line. You validate each step. No prompting.
- Model agnostic
- Every edit logged
- Configurable to your credit policy
Functional advisors
Senior RMs from India's leading banksBuilt around Indian credit practice: CMA formats, RBI norms and local filings.
Technology advisors from
GoogleMorgan Stanley
Four steps. You sign off on each.
Endless promptingFour guided steps, no prompt boxOutput you can't checkA source behind every lineA document you don't understandReasoning you can follow
2
Financials (CMA)
Apprentice
Fills Form 2 and Form 3, with a source and a reason for every line.
You
Validate, and edit or reclassify anything with a reason.
3
Credit appraisal
Apprentice
Drafts the borrower and promoter profiles, financial analysis, and group and transaction structure.
You
Review each section, edit it, or ask questions of the data.
4
Full CAM
Apprentice
Puts every validated section into one memo.
You
Read it through, confirm, and download it as a Word file.
Built the way an RM checks a file.
Every number shows where it came from.
Click any line to see its source, why it was classified there, and who changed it.
- Page-level citations
- A reason for each classification
- A notes view to check every note
- Edits and exclusions logged with a reason
| Form 3, ₹ crore | FY24 | FY25 |
|---|---|---|
| EBITDA | 111.7 | 96.1 |
ReconciledEdited by RMSent to the RM to resolve
It finds the reason, not a restatement.
Apprentice follows a change back to its cause across the borrower's, promoter's and guarantor's documents, and cites each step.
- Reads several documents for one conclusion
- Goes to second and third order causes
- Lets you question any section and update it
A general AI tool says: “EBITDA fell because revenue fell.”
- 1EBITDA fell 14% to ₹96 crore.Audited financials, page 41
- 2Revenue fell 9% while fixed costs held.Annual report, page 112
- 3FY24 included a one-time ₹38 crore export subsidy. Without it, margin rose 60 bps.Annual report note 24, and earnings presentation slide 7
A first draft of the group structure.
The borrower sits at the centre with every subsidiary mapped. Overseas entities are shown in a different colour. You correct it rather than draw it.
- Holdings and guarantors included
- Transaction flow from the term sheet
- Edit and save in place
What we measure.
How often it's right, and how long it takes to reach a signed-off result.
CMA
Financial model, Form 2 and Form 3
AccuracyLine items reconciled to source
100%
Agent run time~17 min
RM validation time~10 min
CAM
Credit appraisal, reasoned sections
AccuracyLines the RM accepts as written
92%
Agent run time~25 min
RM validation time~30 min
Tested on 15 sample credit files. Agents run on Gemini 3.7 Flash and Gemini 3.5 Flash Lite.
Ready for your risk and audit teams.
Full audit trail
Every edit and reclassification is recorded with its reason, so any CAM can be reproduced later.
Works with any model
Results are consistent across model providers, so there's no vendor lock-in.
Set up for your credit policy
Sections, templates and rules follow how your bank assesses credit.
Aligned with RBI's FREE-AI framework
Citations, reasons and the edit log make each AI-assisted decision explainable.
“Two credit memos a week, each built from about 5,000 pages on the borrower and its guarantors. Something that mattered was always going to slip through. And every hour in that paperwork was an hour away from my customers, which was the job I was actually there to do. So I built Apprentice.”
Bring a credit file. We'll build the CAM with you.
In a walkthrough, we run a real file and show the source behind any line you want to question.
© Apprentice Systems. Made in India.
Company names, figures and documents on this site are synthetic. No client data is used.