Skip to content

rmn2178/Intelli-credit

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🏦 Intelli-Credit — AI-Powered Credit Decisioning Engine

Bank-grade credit intelligence. Automated. Auditable. Agentic.


💡 Core Idea

Intelli-Credit is an autonomous, multi-agent AI system that transforms the way banks and NBFCs evaluate loan applications. It ingests raw financial documents — balance sheets, GST returns, ITRs, bank statements — and produces a fully reasoned, evidence-cited Credit Appraisal Memorandum (CAM) with fraud screening, regulatory compliance checks, and a simulated stress test, all without human intervention until the final decision gate.


⚠️ The Problem

Credit underwriting in India is slow, inconsistent, and dangerously manual.

A typical SME loan application sits on a credit analyst's desk for 10–21 working days. The analyst manually extracts numbers from PDFs, builds ratios in spreadsheets, cross-checks GST filings against P&L statements, and writes a CAM from scratch — every single time. The process is:

  • Opaque — decisions rely on analyst intuition with little documented reasoning
  • Inconsistent — two analysts evaluating the same file often reach different conclusions
  • Slow — multi-week turnarounds kill SME cash flow and borrower trust
  • Fraud-blind — manual review misses sophisticated document manipulation and GST mismatches
  • Non-compliant by default — RBI and Basel III ratio checks are often done informally or skipped

🔥 Effects of the Problem

Impact Area Real-World Effect
SMEs Loan rejections due to poor documentation, not poor creditworthiness
Banks NPA risk from approving loans based on fabricated financials
Analysts Burnout from repetitive, low-value extraction tasks
Regulators Audit trails are weak; CAMs rarely cite their evidence sources
Economy Credit gap for India's 63 million MSMEs exceeds ₹25 lakh crore

✅ The Solution

Intelli-Credit deploys a pipeline of specialized AI agents, each owning a distinct slice of the underwriting workflow:

  1. Document Routing Agent — classifies uploaded files (structured PDFs, scanned images, unstructured text) and routes them to the right parser
  2. OCR Engine — extracts text from scanned documents with layout awareness
  3. Financial Agent — parses balance sheets and P&L statements into structured financial data
  4. GST Agent — reconciles GSTR-1 vs P&L, GSTR-2A vs GSTR-3B, flagging turnover mismatches
  5. Fraud Radar Agent — runs OSINT collection, cross-references registry data (CIN/GSTIN), and scores fraud probability
  6. Forensic Audit Agent — detects document tampering, round-tripping, and anomalies using a causal reasoning engine
  7. Verification Agent — validates all evidence items, classifying them as Green / Yellow / Red
  8. CAM Agent — generates a bank-grade Credit Appraisal Memorandum with formula citations and source references
  9. Committee Agent — runs a three-persona AI debate (Risk Officer, Compliance Officer, Business Officer) to simulate credit committee review
  10. Regulatory Engine — checks RBI exposure norms, DSCR thresholds, and Basel III capital ratio compliance
  11. Simulation Engine — stress-tests the borrower across worst-case revenue, rate, and expense scenarios

The result: a complete, auditable credit decision — in minutes, not weeks.


🆚 Existing Solutions vs. Intelli-Credit Innovations

Dimension Existing Tools Intelli-Credit
CAM Generation Analyst writes manually in Word AI generates structured CAM with [Source: Doc, Section] and [Logic: Formula] citations
Fraud Detection Rule-based flag lists Multi-signal fraud radar: OSINT + GST reconciliation + forensic document analysis + causal engine
Credit Committee Human meeting, unrecorded debate Autonomous 3-persona AI committee with full transcript
Stress Testing Static sensitivity table Monte Carlo-style simulation across revenue, rate, and cost shock scenarios
Regulatory Checks Post-approval compliance review Real-time RBI/Basel III compliance enforced during underwriting
Evidence Auditability Analyst notes in margin Every claim tagged Green/Yellow/Red with traceable evidence trail
Document Handling Analyst reads PDFs manually Automated routing: OCR for scans, structured parser for digitals, LLM for unstructured text
SMT Validation None Satisfiability Modulo Theory validator cross-checks financial logical consistency
Knowledge Graph None Entity relationship graph links company, directors, GST, and registry data
Federated Privacy Centralized sensitive data Federated learning stub isolates sensitive borrower data per institution

🛠️ Tech Stack

Backend

Layer Technology
Runtime Python 3.11+
API Framework FastAPI
Agent Orchestration Custom async multi-agent pipeline (asyncio, AsyncGenerator)
LLM Provider Ollama (local) — llama3.2, lfm2.5-thinking; optional OpenAI / Anthropic / Google APIs
OCR Custom OCR engine (core/ocr.py)
Knowledge Graph core/knowledge_graph.py — entity relationship mapping
Causal Reasoning core/causal_engine.py — anomaly and causation detection
SMT Validation core/smt_validator.py — logical consistency verification
OSINT Collection core/osint_collector.py + core/web_search.py
Regulatory Engine core/regulatory_engine.py — RBI / Basel III checks
Forensic Engine core/forensic_engine.py — document integrity analysis
Session Store models/database.py — in-memory session management
Data Schemas Pydantic (models/schemas.py)
Testing Pytest

Frontend

Layer Technology
Framework Next.js 14 (App Router)
Language TypeScript
Styling Tailwind CSS + custom global CSS design system
UI Pages Upload → Processing → Intelligence → Results → Regulatory → Simulator
Fonts Inter, Public Sans (body) · JetBrains Mono (data/code)

Infrastructure & Config

Layer Technology
Environment .env with OLLAMA_BASE_URL, optional cloud API keys
API Bridge NEXT_PUBLIC_API_URL connects Next.js frontend to FastAPI backend
Package Management pip (backend) · npm (frontend)

🗂️ Project Structure

./
├── backend/
│   ├── agents/          # Specialized AI agents (CAM, GST, Fraud, Committee, etc.)
│   ├── core/            # Engines: OCR, LLM client, Forensic, Regulatory, OSINT, Knowledge Graph
│   ├── models/          # Database session store and Pydantic schemas
│   ├── tests/           # Pytest test suite
│   ├── main.py          # FastAPI application entry point
│   └── requirements.txt
└── frontend/
    ├── app/             # Next.js App Router pages
    │   ├── upload/      # Document upload
    │   ├── processing/  # Live agent processing stream
    │   ├── intelligence/# Analysis results
    │   ├── results/     # CAM output
    │   ├── regulatory/  # Compliance dashboard
    │   └── simulator/   # Stress test simulator
    ├── styles/          # Global CSS design system
    └── next.config.js

🚀 Quick Start

# 1. Clone and configure
cp .env.example .env

# 2. Start Ollama and pull models
ollama pull llama3.2:latest

# 3. Run backend
cd backend
pip install -r requirements.txt
uvicorn main:app --reload

# 4. Run frontend
cd frontend
npm install
npm run dev

Open http://localhost:3000 — upload your financial documents and let the agents work.


Built for Indian banking. Designed to RBI standards. Evidence-first by default.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors