FinRegent

Agentic AI for Financial Governance & Compliance

Multi-agent reasoning with real-time regulatory intelligence for financial institutions

FinRegent is an enterprise-grade AI compliance platform that transforms how financial institutions manage regulatory requirements. Built on an Agentic RAG Framework, it provides autonomous reasoning capabilities to help compliance teams navigate complex regulations from RBI, SEBI, and other regulatory bodies with confidence and transparency.

Why FinRegent?

Financial compliance is complex, time-consuming, and constantly evolving. FinRegent simplifies this challenge by:

Understanding Context: Ask questions in natural language and receive precise, contextual answers backed by regulatory sources
Autonomous Reasoning: Multi-agent system that can analyze, cross-reference, and provide explainable compliance guidance
Real-time Intelligence: Automatic monitoring and integration of regulatory updates from multiple sources
Unified Knowledge: Seamlessly combines RBI guidelines, SEBI requirements, and internal policies in one intelligent system

Features

🤖 AI-Powered Compliance Assistant

Natural language chat interface powered by Lyzr AI agents trained on financial regulations. Get instant answers to compliance questions with:

Contextual Understanding: Multi-turn conversations that maintain context across sessions
Source Attribution: Every response includes citations and confidence scores
Voice Interaction: Hands-free compliance queries with voice input/output
Streaming Responses: Real-time answer generation with rich markdown formatting

📚 Intelligent Document Management

Centralized knowledge base for all regulatory documents:

Auto-Classification: Automatic categorization by source (RBI, SEBI, Internal) and category
Smart Search: Tag-based filtering and full-text search across all documents
AI Processing: Automated summarization and key point extraction
Version Control: Track document changes and maintain compliance history

🔄 Automated Regulatory Updates

Stay current with evolving regulations:

Scheduled Monitoring: Automated bi-weekly checks for regulatory updates
Link Extraction: AI agent extracts relevant links from regulatory websites
Knowledge Base Training: Automatically trains RAG system with new information
Email Notifications: Get notified when your knowledge base is updated

🔐 Enterprise Security

Built with security and compliance in mind:

Clerk Authentication: Enterprise-grade SSO and user management
API Signature Verification: Cryptographic signatures for all API requests
Webhook Integration: Real-time user synchronization with audit trails
Role-Based Access: Granular permission controls (coming soon)

🎨 Modern User Experience

Beautiful, responsive interface that works everywhere:

Dark Mode Support: Full theme customization with persistent preferences
Responsive Design: Optimized for desktop, tablet, and mobile
Accessibility First: WCAG 2.1 compliant UI components
Real-time Updates: Live activity tracking with visual timeline

Quick Start

Prerequisites

Node.js 18.x or higher
MongoDB database (local or MongoDB Atlas)
Clerk account for authentication
Lyzr AI account and API credentials

Installation

Clone the repository:

git clone https://github.com/AJAmit17/NH-SJC25HACK-GenAI-026-Team-UDAAN.git
cd NH-SJC25HACK-GenAI-026-Team-UDAAN

Install dependencies:

npm install

Configure environment variables:

Create a .env file in the root directory:

# Clerk Authentication
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_test_...
CLERK_SECRET_KEY=sk_test_...
CLERK_WEBHOOK_SECRET=whsec_...

# Database
DATABASE_URL="mongodb+srv://..."

# Lyzr AI
LYZR_API_KEY=sk-default-...
LYZR_AGENT_ID=...
LYZR_RAG_ID=...
LYZR_UPDATES_AGENT_ID=...
LYZR_LINK_EXTRACTION_AGENT_ID=...
LYZR_AUTO_RAG_ID=...

# Security
API_SIGNATURE_SECRET=your_random_secret_min_32_chars

# Inngest (for scheduled jobs)
INNGEST_SIGNING_KEY=...
INNGEST_EVENT_KEY=...

# Email Notifications
EMAIL_HOST=smtp.gmail.com
EMAIL_PORT=587
EMAIL_SECURE=false
EMAIL_USER=your-email@gmail.com
EMAIL_PASS=your-app-password

Initialize the database:

npx prisma generate
npx prisma db push

Start the development server:

npm run dev

Visit http://localhost:3000 to see your application.

Architecture

Architecture Diagram

Client Layer: Next.js UI with Clerk-secured access.
Services Layer: Lyzr agents, Prisma API, and signature verification.
Data & Jobs: MongoDB storage plus Inngest-driven regulatory updates.

Tech Stack

Category	Technologies
Framework	Next.js 15 (App Router), React 19
Language	TypeScript
Database	MongoDB + Prisma ORM
Authentication	Clerk
AI/ML	Lyzr AI Agents
Styling	Tailwind CSS, Shadcn UI
State Management	TanStack Query (React Query)
Background Jobs	Inngest
Email	Nodemailer
Real-time	Server-Sent Events (SSE)

Key Components

Multi-Agent System

FinRegent uses a specialized multi-agent architecture:

Chat Agent: Main conversational interface for compliance queries
Link Extraction Agent: Automatically discovers relevant regulatory documents
Updates Agent: Monitors and processes regulatory changes
RAG Agent: Retrieves and augments responses with knowledge base context

API Security Layer

All API requests use cryptographic signature verification:

// Client generates signature
const signature = await generateSignatureAction(JSON.stringify(payload));

// Server verifies signature
const isValid = verifyApiSignature(payload, signature);

This ensures request integrity and prevents unauthorized API access.

Automated Training Pipeline

User Enables Cron Job
       ↓
Inngest Scheduled Function (Every 2 weeks)
       ↓
Link Extraction Agent → Discovers new regulatory documents
       ↓
RAG Agent → Trains knowledge base with new content
       ↓
Email Notification → User receives update summary

Project Structure

finregent/
├── prisma/
│   └── schema.prisma              # Database schema (User, Document, Chat, etc.)
├── src/
│   ├── actions/
│   │   ├── api-key.ts            # Server action to fetch API keys
│   │   └── generate-signature.ts  # Cryptographic signature generation
│   ├── app/
│   │   ├── (auth)/               # Public authentication routes
│   │   ├── (routes)/             # Protected application routes
│   │   │   ├── workspace/        # Main chat workspace
│   │   │   ├── documents/        # Document management
│   │   │   ├── settings/         # User preferences
│   │   │   └── updates/          # Regulatory updates feed
│   │   └── api/
│   │       ├── lyzr/chat/        # Chat API with signature verification
│   │       ├── users/webhook/    # Clerk user synchronization
│   │       └── inngest/          # Background job endpoints
│   ├── components/
│   │   ├── chat.tsx              # Main chat interface
│   │   ├── sidebar/              # Navigation components
│   │   ├── shared/               # Reusable components (Timeline, VoiceChat)
│   │   └── ui/                   # Shadcn UI library (50+ components)
│   ├── lib/
│   │   ├── lyzrService.ts        # Lyzr AI integration
│   │   ├── security.ts           # Signature & webhook verification
│   │   ├── prisma.ts             # Database client
│   │   ├── email.ts              # Email service
│   │   └── inngest/              # Background job definitions
│   └── middleware.ts             # Auth & route protection
└── package.json

Configuration

Setting up Clerk

Create an application at clerk.com
Configure sign-in options (Email, Google, etc.)
Copy API keys to .env
Add webhook endpoint:
- URL: https://your-domain.com/api/users/webhook
- Events: user.created, user.updated, user.deleted

Setting up Lyzr AI

Sign up at lyzr.ai
Create AI agents for:
- Main Chat Agent: Compliance Q&A
- Link Extraction Agent: Document discovery
- Updates Agent: Regulatory monitoring
- RAG Agent: Knowledge base management
Train agents with your regulatory documents
Copy agent IDs to .env

Setting up Automated Updates

Note

The automated training pipeline requires Inngest for scheduled jobs.

Create account at inngest.com
Create a new app and copy keys to .env
Users can enable/disable automated updates in Settings
System runs bi-weekly (configurable in src/lib/inngest/functions.ts)

Development

Available Scripts

Command	Description
`npm run dev`	Start development server with Turbopack
`npm run build`	Build production bundle
`npm run start`	Start production server
`npm run lint`	Run ESLint
`npx prisma studio`	Open database GUI
`npx prisma db push`	Push schema changes to DB

API Endpoints

`POST /api/lyzr/chat`

Send a message to the AI compliance assistant.

Request:

{
  "message": "What are the KYC requirements for banks?",
  "sessionId": "optional-session-id",
  "agentId": "your-agent-id",
  "userId": "user-id",
  "signature": "generated-signature",
  "assets": ["optional-file-ids"]
}

Response:

{
  "response": "AI response content...",
  "session_id": "session-id",
  "user_id": "user-id"
}

`POST /api/users/webhook`

Clerk webhook endpoint for user synchronization.

Events handled:

user.created - Creates user in MongoDB
user.updated - Updates user information
user.deleted - Removes user and associated data

`POST /api/lyzr/upload`

Upload documents to Lyzr AI for processing.

Request: Multipart form data with files

Response:

{
  "assets": [
    {
      "asset_id": "...",
      "filename": "document.pdf"
    }
  ]
}

Deployment

Deploy to Vercel

Push your code to GitHub
Import repository in Vercel
Add environment variables
Deploy

Important

Update Clerk webhook URL to your production domain after deployment.

Environment Checklist

Ensure all variables are set:

✅ Clerk credentials (publishable key, secret, webhook secret)
✅ MongoDB connection string
✅ All Lyzr AI agent IDs and API key
✅ API signature secret (min 32 characters)
✅ Inngest keys (for automated updates)
✅ Email configuration (for notifications)

Troubleshooting

Database Connection Issues

Verify MongoDB connection string format
Check IP whitelist in MongoDB Atlas
Ensure database user has read/write permissions
Run npx prisma generate to regenerate client

Authentication Errors

Verify Clerk API keys in .env
Check that redirect URLs match your domain
Ensure webhook endpoint is accessible
Clear browser cookies and localStorage

AI Not Responding

Verify Lyzr API key and agent IDs
Check that agents are properly trained
Review browser network tab for API errors
Ensure signature generation is working (check console)

Build Errors

Delete .next folder and node_modules
Run npm install fresh
Check for TypeScript errors with npm run lint
Ensure all environment variables are set

Resources

Built with ❤️ by Team UDAAN

GitHub • Powered by Lyzr AI

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
.github/assets		.github/assets
prisma		prisma
public		public
src		src
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
components.json		components.json
eslint.config.mjs		eslint.config.mjs
next.config.ts		next.config.ts
package-lock.json		package-lock.json
package.json		package.json
postcss.config.mjs		postcss.config.mjs
tsconfig.json		tsconfig.json

AJAmit17/FinRegent

Folders and files

Latest commit

History

Repository files navigation

FinRegent

Why FinRegent?

Features

🤖 AI-Powered Compliance Assistant

📚 Intelligent Document Management

🔄 Automated Regulatory Updates

🔐 Enterprise Security

🎨 Modern User Experience

Quick Start

Prerequisites

Installation

Architecture

Architecture Diagram

Tech Stack

Key Components

Multi-Agent System

API Security Layer

Automated Training Pipeline

Project Structure

Configuration

Setting up Clerk

Setting up Lyzr AI

Setting up Automated Updates

Development

Available Scripts

API Endpoints

POST /api/lyzr/chat

POST /api/users/webhook

POST /api/lyzr/upload

Deployment

Deploy to Vercel

Environment Checklist

Troubleshooting

Resources

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Languages

`POST /api/lyzr/chat`

`POST /api/users/webhook`

`POST /api/lyzr/upload`

Packages