🏆 FinSight Copilot - Due Diligence AI Assistant

Real-time Covenant Monitoring with AI-Powered Document Extraction

An intelligent financial agent that monitors loan covenants in real-time, extracting data from complex financial documents and alerting on breaches instantly.

🎯 Problem Statement

Investors, M&A professionals, and finance leaders need to monitor covenant compliance across evolving data rooms. Manual tracking is slow, error-prone, and misses critical breaches.

Our Solution: An AI-powered copilot that:

✅ Extracts covenant terms from PDFs, images, and tables using LandingAI DPT-2
✅ Monitors financial data in real-time using Pathway framework
✅ Detects breaches instantly with red flag alerts
✅ Sends automated email notifications
✅ Generates living memos that auto-update
✅ Provides conversational Q&A interface

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    DATA ROOM (Live Files)                   │
│  PDF │ JSON │ Images │ CSV │ HTML/iXBRL                     │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│              PATHWAY LIVE FILE MONITOR                      │
│  (Watches for new/modified files - instant processing)      │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│           LANDINGAI ADE (DPT-2) EXTRACTION                  │
│  • Covenant terms from PDFs/Images                          │
│  • Financial tables with high accuracy                      │
│  • Parallel extraction for speed                            │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│              COVENANT COMPLIANCE ENGINE                     │
│  • Leverage Ratio Check                                     │
│  • Interest Coverage Check                                  │
│  • Liquidity Check                                          │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│                  INTELLIGENT OUTPUTS                        │
│  📧 Email Alerts  │  📝 Living Memos  │  💬 Chatbot Q&A    │
└─────────────────────────────────────────────────────────────┘

🚀 Tech Stack

Core Technologies (Required)

LandingAI ADE (DPT-2): Document extraction with vision capabilities
Pathway Framework: Live file monitoring and real-time indexing

Supporting Stack

Python 3.10+: Core language
PyPDF2: PDF processing and truncation
Watchdog: File system monitoring
SMTP: Email alert delivery
FastAPI: REST API (optional)

📁 Project Structure

FinSight_Copilot/
├── hackathon_copilot.py          # Main copilot implementation
├── pathway_monitor.py             # Live file monitoring
├── due_diligence_copilot (1).py  # Original prototype
├── .env                           # Environment configuration
├── covenant-copilot/
│   ├── data/                      # Data room
│   │   ├── JSON/file.json        # SEC financial data
│   │   ├── PDF/file.pdf          # Credit agreements
│   │   ├── Image/file.jpeg       # Covenant tables
│   │   ├── CSV/                  # Normalized data
│   │   └── HTML : iXBRL/         # SEC filings
│   ├── apps/
│   │   └── api/                  # FastAPI backend
│   └── memos/                    # Generated memos
└── HACKATHON_README.md           # This file

🔧 Setup & Installation

1. Clone Repository

git clone https://github.com/kushal511/FinSight_Copilot.git
cd FinSight_Copilot

2. Create Virtual Environment

python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install Dependencies

pip install --upgrade pip
pip install pandas beautifulsoup4 lxml requests PyPDF2 pillow pdf2image \
            watchdog python-dotenv landingai_ade fastapi uvicorn

4. Configure Environment

Create .env file:

# LandingAI API Key (get from https://va.landing.ai/)
LANDING_AI_API_KEY=your_api_key_here

# Data paths
LOCAL_DATA_PATH=./covenant-copilot/data

# Email alerts (optional)
SMTP_SERVER=smtp.gmail.com
SMTP_PORT=587
EMAIL_FROM=alerts@covenantcopilot.ai
EMAIL_PASSWORD=your_app_password
ALERT_RECIPIENTS=analyst@company.com,manager@company.com

5. Prepare Data Room

Ensure these files exist in covenant-copilot/data/:

JSON/file.json - SEC CompanyFacts API data
PDF/file.pdf - Credit agreement (auto-truncated to 50 pages)
Image/file.jpeg - Covenant table screenshot
CSV/file.csv - Normalized financial metrics

🎬 Demo Execution

Option 1: Full Demo (All Features)

python hackathon_copilot.py

What it does:

✅ Extracts covenant terms using LandingAI DPT-2
✅ Analyzes financial data from SEC JSON
✅ Runs covenant compliance checks
✅ Detects red flags and breaches
✅ Sends email alerts (or saves to file)
✅ Generates living memo (Markdown + HTML)
✅ Runs conversational chatbot demo

Expected Output:

🏆 HACKWITHBAY 2025 - DUE DILIGENCE COPILOT
   Track 1: Covenant Monitoring with Live Updates
================================================================================

📋 STEP 1: Document Extraction (LandingAI DPT-2)
--------------------------------------------------------------------------------
🖼️  Extracting from Image: file.jpeg
  ✅ Extracted 7 chunks

📊 STEP 2: Financial Data Extraction
--------------------------------------------------------------------------------
📊 Reading SEC JSON: file.json
  ✅ Debt=$4994M, EBITDA=$1322M, Cash=$15587M

🔍 STEP 3: Covenant Compliance Check
--------------------------------------------------------------------------------
Company: Tesla Inc.
Period: Q2 2025
Status: 🔴 BREACH

❌ LEVERAGE: 3.78 (limit: 3.5) - BREACH
✅ COVERAGE: 8.16 (limit: 2.0) - PASS
✅ LIQUIDITY: 15587.0 (limit: 100.0) - PASS

📧 STEP 4: Email Alert System
--------------------------------------------------------------------------------
📧 Sending breach alert to 1 recipients...
  ✅ Alert saved to: covenant-copilot/data/breach_alert_20251004_183331.html

📝 STEP 5: Living Memo Generation
--------------------------------------------------------------------------------
📝 Generating living memo...
  ✅ Memo saved to: covenant-copilot/data/memos/covenant_memo_20251004_183331.md
  ✅ HTML version: covenant-copilot/data/memos/covenant_memo_20251004_183331.html

💬 STEP 6: Conversational Chatbot Demo
--------------------------------------------------------------------------------
❓ Q: What is the overall covenant status?
💡 A: The overall covenant status for Tesla Inc. is **BREACH**. There are 1 red flag(s) detected.

❓ Q: What is the leverage ratio?
💡 A: The leverage ratio is 3.78x against a limit of 3.5x. Status: BREACH. Margin: -0.28x.

✅ DEMO COMPLETE - All Components Working

Option 2: Pathway Live Monitoring

# Monitor for 30 seconds
python pathway_monitor.py 30

# Monitor indefinitely (Ctrl+C to stop)
python pathway_monitor.py

What it does:

🔍 Scans existing files in data room
👀 Watches for new/modified files
⚡ Processes changes instantly
🚨 Triggers alerts on breaches
📝 Updates living memos automatically

Demo Scenario:

Start monitoring: python pathway_monitor.py

In another terminal, update a file:

# Simulate new financial data
cp covenant-copilot/data/JSON/file.json covenant-copilot/data/JSON/file_updated.json

Watch instant processing and alerts!

Option 3: FastAPI Server

cd covenant-copilot/apps/api
venv/bin/uvicorn main:app --reload --port 8000

Access API docs: http://localhost:8000/docs

📊 Key Features & Differentiators

1. LandingAI DPT-2 Integration ✨

Accurate table extraction from covenant documents
Parallel processing for faster parsing
Vision capabilities for scanned documents
Automatic PDF truncation to 50 pages for API limits

2. Pathway Live Monitoring 🔄

Real-time file watching with instant processing
Hybrid indexing (ready for vector + BM25)
Event-driven architecture for scalability
Automatic re-processing on file updates

3. Intelligent Alerting 🚨

Red flag detection with severity levels
Email notifications with HTML formatting
Evidence-based alerts with source citations
Configurable thresholds per covenant type

4. Living Memo 📝

Auto-updating documentation on every check
Markdown + HTML formats for flexibility
Executive summary with actionable insights
Audit trail with timestamps

5. Conversational Interface 💬

Natural language Q&A about covenant status
Context-aware responses with financial data
Conversation history tracking
Extensible to LLM for advanced queries

📈 Results & Outputs

Generated Artifacts

covenant_results.json - Full compliance check results
red_flags.json - Detected breaches with severity
covenant_memo_*.md - Living memo (Markdown)
covenant_memo_*.html - Living memo (HTML)
breach_alert_*.html - Email alert (HTML)
chatbot_conversation.json - Q&A history
pathway_results/check_*.json - Live monitoring results

Sample Output: Red Flags

[
  {
    "rule": "leverage",
    "actual": 3.78,
    "limit": 3.5,
    "margin": 0.28,
    "severity": "HIGH",
    "message": "🚨 LEVERAGE BREACH: 3.78x exceeds 3.5x limit",
    "timestamp": "2025-10-04T18:33:31.481529"
  }
]

📚 References

LandingAI Resources

Playground: https://va.landing.ai/
Developer Docs: https://docs.landing.ai/ade/ade-overview
Python Library: https://github.com/landing-ai/ade-python
Discord: https://discord.com/invite/RVcW3j9RgR

Pathway Resources

Documentation: https://pathway.com/developers/
GitHub: https://github.com/pathwaycom/pathway
RAG 101 Bootcamp: https://pathway.com/developers/showcases/

🎯 Future Enhancements

LLM Integration: Advanced conversational AI with GPT-4
Vector Search: Hybrid indexing for intelligent document search
Multi-Company Support: Monitor multiple portfolios simultaneously
Mobile Applications: iOS/Android apps for on-the-go monitoring
Enterprise Integrations: Slack, Teams, Bloomberg Terminal connectivity

📄 License

MIT License - Built for HackWithBay 2025

🚀 Ready to revolutionize covenant monitoring!

Name		Name	Last commit message	Last commit date
Latest commit History 9 Commits
FinSight_Copilot		FinSight_Copilot
README.md		README.md

Folders and files

Latest commit

History

Repository files navigation

🏆 FinSight Copilot - Due Diligence AI Assistant

🎯 Problem Statement

🏗️ Architecture

🚀 Tech Stack

Core Technologies (Required)

Supporting Stack

📁 Project Structure

🔧 Setup & Installation

1. Clone Repository

2. Create Virtual Environment

3. Install Dependencies

4. Configure Environment

5. Prepare Data Room

🎬 Demo Execution

Option 1: Full Demo (All Features)

Option 2: Pathway Live Monitoring

Option 3: FastAPI Server

📊 Key Features & Differentiators

1. LandingAI DPT-2 Integration ✨

2. Pathway Live Monitoring 🔄

3. Intelligent Alerting 🚨

4. Living Memo 📝

5. Conversational Interface 💬

📈 Results & Outputs

Generated Artifacts

Sample Output: Red Flags

📚 References

LandingAI Resources

Pathway Resources

🎯 Future Enhancements

📄 License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages