Skip to content

Laya-Myadam/VaultIQ

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
backend
backend
 
 
frontend
frontend
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 

Repository files navigation

VaultIQ — AI Banking Intelligence Platform

Full-stack AI platform for financial institutions — fraud detection, credit underwriting, AML screening, regulatory compliance, risk stress testing, document intelligence, and automated report generation.

Live Gateway: https://vaultiq-frontend-934314486311.us-central1.run.app/

What Is VaultIQ?

VaultIQ is an AI-powered banking operating system that brings together multiple intelligence modules under one platform. It replaces hours of manual analyst work with real-time AI decisions — scoring transactions for fraud in milliseconds, underwriting loan applications instantly, screening customers against AML/KYC watchlists, and drafting regulatory filings like SARs and CTRs in seconds.

Each module is powered by a LangGraph agent pipeline backed by Groq's LLaMA 3.3 70B model, with XGBoost ML models for fraud and credit scoring.

Screenshots

Fraud Intel · Credit Scoring · Compliance · Risk · AML · Report Studio · Loan Monitor · Doc AI

Modules

Module What It Does
Fraud Intel Real-time transaction risk scoring. Inputs: amount, hour, frequency, distance, origin. Outputs: fraud score 0–100, APPROVE/REVIEW/BLOCK decision, behavioral explanation
Credit AI loan underwriting. Scores applications using XGBoost + LangGraph. Returns approval decision, suggested credit limit, risk explanation
Compliance RAG-based regulatory assistant. Answers questions on BSA, AML, KYC, OFAC, GLBA, FCRA, TILA, CRA. Also screens transactions for compliance flags
Risk Portfolio stress testing across 4 scenarios — mild recession, severe recession, interest rate shock, cyber attack. Computes NPA ratio, Tier 1 capital, liquidity, LTD ratio
Loan Monitor Real-time portfolio health analysis. Enter portfolio metrics + individual watch-list loans. Returns health score, early warning indicators, per-loan MONITOR/WATCH/RESTRUCTURE decisions
AML Intel NLP transaction narrative analysis — extracts entities, detects AML typologies (structuring, layering, smurfing). Customer screening against OFAC/PEP/sanctions with risk score
Report Studio AI-generated banking reports: SAR, CTR, Risk Assessment, Credit Portfolio Summary, Compliance Audit, Board Executive Summary
Doc AI Upload PDF/DOCX/TXT → instant AI summary + conversational Q&A on the document content

Tech Stack

Layer Technology
Frontend React 19 + Vite 8
Styling CSS variables + inline styles — custom light/ice design system
Fonts Space Grotesk (display) · Outfit (body) · DM Mono (data)
Routing React Router v7
Charts Recharts
Icons Lucide React
Frontend Deploy Firebase Hosting
API Gateway Node.js + Express
Gateway Deploy Google Cloud Run
AI Services Python + FastAPI + Uvicorn
LLM Groq — llama-3.3-70b-versatile
Agent Orchestration LangGraph (multi-step agent pipelines)
ML Models XGBoost + scikit-learn (fraud scoring, credit scoring)
Vector DB ChromaDB (document embeddings for Doc AI)
PDF Parsing pdfplumber + python-docx
AI Deploy Google Cloud Run

Project Structure

ai-banking-platform/
│
├── frontend/                          # React + Vite SPA
│   ├── src/
│   │   ├── App.jsx                    # Router + auth guard (PrivateRoute)
│   │   ├── index.css                  # Design tokens, fonts, dot-grid texture
│   │   ├── pages/
│   │   │   ├── Dashboard.jsx          # KPI overview, live alerts, trend charts
│   │   │   ├── Fraud.jsx              # Transaction risk scoring
│   │   │   ├── Credit.jsx             # Loan underwriting
│   │   │   ├── Compliance.jsx         # Regulatory Q&A + transaction check
│   │   │   ├── Risk.jsx               # Portfolio stress testing
│   │   │   ├── LoanMonitor.jsx        # Loan portfolio health
│   │   │   ├── AML.jsx                # Narrative NLP + customer screening
│   │   │   ├── Reports.jsx            # Report Studio (SAR, CTR, etc.)
│   │   │   ├── Documents.jsx          # Doc upload + AI Q&A
│   │   │   ├── Login.jsx              # Auth (localStorage → Firebase later)
│   │   │   └── Settings.jsx           # Profile, preferences, session
│   │   ├── components/
│   │   │   ├── layout/
│   │   │   │   ├── Sidebar.jsx        # Icon-only nav (68px), tooltip labels
│   │   │   │   └── Topbar.jsx         # Frosted glass header, live clock
│   │   │   ├── ui/
│   │   │   │   ├── PageLayout.jsx     # Wraps Topbar + page content
│   │   │   │   └── FormCard.jsx
│   │   │   └── cards/
│   │   │       ├── MetricCard.jsx
│   │   │       ├── AlertCard.jsx
│   │   │       └── ModuleCard.jsx
│   │   └── services/
│   │       └── api.js                 # All API calls — maps frontend → backend fields
│   ├── .env                           # VITE_API_URL
│   └── Dockerfile
│
├── backend/
│   ├── api-gateway/                   # Express proxy — routes frontend → AI services
│   │   ├── src/
│   │   │   ├── index.js               # Route registration + health check
│   │   │   └── routes/
│   │   │       ├── fraud.js
│   │   │       ├── credit.js
│   │   │       ├── compliance.js
│   │   │       ├── risk.js
│   │   │       ├── docs.js
│   │   │       ├── reports.js
│   │   │       ├── aml.js
│   │   │       └── loans.js
│   │   └── Dockerfile
│   │
│   └── ai-services/                   # FastAPI — all AI logic
│       ├── run.py                     # Unified entry point (all routes, port 8080)
│       ├── shared/
│       │   └── llm.py                 # get_llm() → Groq ChatGroq instance
│       ├── fraud/
│       │   ├── agent.py               # LangGraph: score_transaction → explain_decision
│       │   ├── model.py               # XGBoost model loader
│       │   └── main.py
│       ├── credit/
│       │   ├── agent.py               # LangGraph: score → limit → explain
│       │   ├── model.py
│       │   └── main.py
│       ├── compliance/
│       │   ├── rag_chain.py           # Transaction compliance checker
│       │   └── main.py
│       ├── risk/
│       │   ├── agent.py               # LangGraph: analyze → stress → recommend
│       │   ├── model.py               # Ratio calculators, stress test functions
│       │   └── main.py
│       ├── docs/
│       │   └── main.py                # PDF/DOCX extract, ChromaDB, RAG Q&A
│       ├── requirements.txt
│       ├── .env                       # GROQ_API_KEY
│       └── Dockerfile
│
└── README.md

API Reference

Gateway base URL

https://vaultiq-gateway-934314486311.us-central1.run.app/api
Method Endpoint Description
POST /fraud/analyze Score a transaction for fraud
POST /fraud/batch Batch-score multiple transactions
POST /credit/evaluate Evaluate a loan application
POST /compliance/query Ask a regulatory question
POST /compliance/check-transaction Screen transaction for compliance
POST /risk/analyze Run portfolio stress test
POST /docs/upload Upload document (multipart)
POST /docs/ask Q&A on uploaded document
GET /health Gateway + AI service health check

Note: /reports, /aml, /loans routes are handled client-side via the deployed /compliance/query endpoint — no separate service needed.

Local Setup

Prerequisites

1 — AI Services

cd backend/ai-services

python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

pip install -r requirements.txt

# Create .env
echo "GROQ_API_KEY=your_key_here" > .env

python run.py                   # → http://localhost:8080

2 — API Gateway

cd backend/api-gateway
npm install
npm start                       # → http://localhost:3001

3 — Frontend

cd frontend
npm install

# For local backend:
echo "VITE_API_URL=http://localhost:3001/api" > .env.local

npm run dev                     # → http://localhost:5173

Deployment

Backend — Google Cloud Run

# AI Services
gcloud run deploy vaultiq-ai \
  --source ./backend/ai-services \
  --region us-central1 \
  --allow-unauthenticated \
  --memory 2Gi --cpu 2 --timeout 300 \
  --set-env-vars "GROQ_API_KEY=your_key"

# API Gateway
gcloud run deploy vaultiq-gateway \
  --source ./backend/api-gateway \
  --region us-central1 \
  --allow-unauthenticated \
  --memory 512Mi --cpu 1 \
  --set-env-vars "AI_URL=https://YOUR_AI_SERVICE_URL"

Frontend — Firebase Hosting

cd frontend

# Set production API URL
echo "VITE_API_URL=https://vaultiq-gateway-934314486311.us-central1.run.app/api" > .env.production

npm run build
firebase deploy --only hosting

Auth

Currently uses localStorage for session persistence (any email + password works). Firebase Authentication is the planned replacement — swap the localStorage calls in Login.jsx and App.jsx with the Firebase SDK when ready.

Known Limitations

  • Doc AI on Cloud Run — doc_cache is in-memory. After a cold start, re-upload is needed. Fallback: Q&A uses the document summary as context automatically.
  • Cold starts — First request after inactivity may take 10–20s on Cloud Run.
  • Scanned PDFs — Text extraction requires a searchable PDF. Use ilovepdf.com to OCR first.
  • Report/AML/Loan routes — These call the existing compliance endpoint rather than dedicated routes, so they work without redeployment.

Roadmap

  • Firebase Authentication (replace localStorage)
  • Persistent document storage (Cloud Storage + Firestore)
  • Streaming LLM responses
  • Batch fraud processing dashboard
  • Real transaction data connectors (Plaid, open banking)
  • Role-based access control (analyst / manager / admin)
  • Mobile app

Built with FastAPI · React · LLaMA 3.3 70B · LangGraph · XGBoost · ChromaDB · Google Cloud

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages