About Intelli·Credit

Built for Indian credit officers

Intelli·Credit is an end-to-end AI-powered credit appraisal engine designed for Indian corporate lending workflows. It processes financial documents, conducts autonomous web research, and generates explainable credit decisions.

Document Intelligence. Upload annual reports, bank statements, GST returns, ITR filings, and scanned pages. The system uses a multi-model OCR pipeline — fitz, pdfplumber, and Qwen2.5-VL — to extract structured data from any format, in 22 Indian languages. Cross-validation catches inconsistencies between GST, bank, and tax return data automatically.

Autonomous Research Agent. A Firecrawl-powered agent searches MCA21 filings, eCourts litigation records, news articles, and promoter intelligence — producing a structured research dossier without any manual lookup. Findings are classified by severity and integrated into the risk assessment.

Scoring & CAM Generation. An XGBoost ML model combined with configurable policy rules produces a credit score, risk grade, recommended loan amount, and interest premium. Every decision is SHAP-explainable. The system then generates a complete 9-section Credit Appraisal Memorandum in Word and PDF.

Architecture Pipeline

1

Multi-Source Data Input

PDFs, bank statements, GST XML, ITR JSON, scanned images

2

Document Processing

fitz + pdfplumber + Qwen2.5-VL OCR for 22 Indian languages

3

Structured Knowledge Store

PostgreSQL for structured data, Qdrant for semantic vectors

4

Web Research Agent

Firecrawl-powered autonomous MCA, eCourts, news, promoter checks

5

Risk Scoring Engine

XGBoost model + hard rejection rules + SHAP explainability

6

CAM Generator

9-section Credit Appraisal Memo in Word/PDF with banking layout

7

Credit Officer Portal

Interactive dashboard with Five Cs radar, stress test, RAG chat

Tech Stack

Next.js 14

Frontend framework (App Router)

FastAPI

Backend API with async endpoints

PostgreSQL

Structured data and company records

Qdrant

Vector store for RAG retrieval

XGBoost

ML risk scoring model

Firecrawl

Web research and scraping

Celery

Async task queue for pipeline

Delta Lake

Data versioning via Databricks

Qwen2.5-VL

OCR for scanned documents

Prefect

Workflow orchestration

Built for Indian workflows

GST/GSTR parsingITR + Form 26ASMCA21 checkseCourts verification

Your documents are processed locally and never stored on external servers. All analysis runs within your own deployment.