All work
Quantisense
Automated credit appraisal that cuts turnaround from a week to minutes: document parsing, fraud detection, stress tests, a research agent and CAM report generation.
- Design and engineering
- 2026
- Built
- MERN
- ML
- EDA
- Python
- LLM
- FastAPI
<10 min
Credit appraisal, down from 5–7 days
92.5%
AI risk analysis accuracy
Problem
Credit appraisal is slow and manual. Documents are read by hand, figures are cross-checked by eye and the appraisal memo is written from scratch, which stretched a single case to 5–7 days.
Approach
Quantisense turns the appraisal into a pipeline: parse the documents, flag fraud signals, stress-test the numbers, research the borrower with an agent and assemble the credit appraisal memo (CAM) automatically.
What it does
- Document parsing. Extracts the figures the analysis needs from the submitted documents.
- Fraud detection. Flags anomalies in the data before they reach the memo.
- Stress tests. Runs the borrower's numbers through adverse scenarios.
- Research and company intelligence. LLM agents gather context on the borrower and check the facts they find.
- Bull and bear debate. Opposing agents argue for and against the loan and a chairman agent weighs them into a decision.
- CAM generation. Produces the credit appraisal memo as a Word document.
Outcome
Appraisal time dropped from 5–7 days to under 10 minutes, with 92.5% accuracy on the AI risk analysis.
