At Shriram Finance, credit analysts prepare a CAM (Credit Appraisal Memo) for every customer with an exposure above Rs. 25 lakh. A critical section of the CAM requires manually listing all active loans from the customer's CIBIL report, including sanction amount, outstanding balance, EMI, overdue, DPD, and lender details.
This is straightforward for a customer with 3-4 loans. But CIBIL reports with 50-100+ loan accounts make this a slow, error-prone, and repetitive task, done 6-7 times per month per branch.
AutoCAM eliminates this manual effort entirely.
Upload a CIBIL PDF and get a structured, formatted Excel file in seconds.
- Extracts all loan accounts automatically (active and closed)
- Covers both CRIF High Mark and TransUnion CIBIL report formats
- Validates extracted data against the report's own summary totals
- Outputs an Excel file in the exact format required for the CAM
- Filter by Active / Closed accounts directly in the app or excel
Live demo: autocam-cibil.streamlit.app
Upload screen
After extraction - borrower profile and key metrics
Account table with Active / Closed filter and Excel download
| Before | After |
|---|---|
| 30-60 min manual data entry per CAM | Under 1 minute |
| Risk of transcription errors | Validated against report totals |
| Only feasible for small CIBIL reports | Handles 100+ account reports |
| Done atleast 6-7 times/month per branch | Same frequency, fraction of the effort |
- Upload a digital CIBIL PDF (CRIF High Mark or TransUnion)
- Click Extract Data
- Review the dashboard - borrower name, CIBIL score, account metrics
- Download the pre-formatted Excel file ready for the CAM
If rule-based extraction doesn't pass validation, the app automatically falls back to a Gemini LLM for correction, without any action needed from the user.
Streamlit · PyMuPDF · openpyxl · LangChain + Google Gemini · pandas
- LLM-powered credit analysis: auto-generate risk observations and key points from the extracted CIBIL data
- Support for scanned/image-based PDFs via OCR
- Multi-borrower batch processing
- Scanned (image-based) PDFs are not supported, only digital CIBIL reports with extractable text
- TransUnion reports have no LLM fallback