AI Credit Dispute Automation Platform

Automated credit report dispute processing with AI. Takes a credit report PDF and outputs:

• Metro 2 file (ready for credit bureau submission)
• Human-readable reports (TXT, HTML, PDF)

Data Residency & Security

What This Software Does:

• ✅ Temp CSV files are overwritten 3x before deletion (cannot be recovered)
• ✅ No data is sent to external servers
• ✅ No telemetry or phone-home features

What Banks Must Implement:

• 🔐 Store Metro 2 output files on encrypted drives
• 🔐 Restrict file access to authorized employees only
• 🔐 Encrypt system swap space
• 🔐 Implement backup encryption for compliance (FCRA, GLBA)
• 🔐 Train employees on proper data handling

Recommended Retention Policy:

• Metro 2 files: Delete after 2 years (or per bank policy)
• Audit logs: Retain for 7 years

Features

Feature	Description
AI Analysis	Uses local DeepSeek model to detect credit report errors
Batch Processing	Watch folder for automatic PDF processing
Web Interface	Self-hosted web UI for drag-and-drop processing
Email Notifications	Send results via email when processing completes
Metro 2 Output	Bureau-ready file format
Human Reports	TXT, HTML, and PDF reports for bank employees
On-Premises	Runs on your infrastructure. Data never leaves your control.

System Requirements

Hardware

Component	Minimum	Recommended
RAM	16GB	32GB
Storage	20GB	50GB SSD
CPU	4-core	8-core
GPU	Optional	NVIDIA with 8GB+ VRAM

Software

• Python 3.10+
• Ollama (for AI model)
• DeepSeek-Coder-V2:16b model

Quick Install

Copy and paste these commands one line at a time:

git clone https://github.com/leonardrougeau-ctrl/credit-dispute-cli.git
cd credit-dispute-cli
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

On Windows, replace source venv/bin/activate with:

venv\Scripts\activate

Install AI Model

Step 1: Install Ollama

Linux (Ubuntu/Debian) - copy and paste:

curl -fsSL https://ollama.com/install.sh | sh

Windows: Download from https://ollama.com/download/windows

macOS:

curl -fsSL https://ollama.com/install.sh | sh

Step 2: Pull the DeepSeek Model

ollama pull deepseek-coder-v2:16b

Step 3: Verify the Model is Ready

ollama list

Step 4: Test the Model

ollama run deepseek-coder-v2:16b "Write a Python function"

Press Ctrl+D to exit when done.

Optional: Keep Model Loaded for Faster Processing

curl http://localhost:11434/api/generate -d '{"model": "deepseek-coder-v2:16b", "keep_alive": "24h"}'

Offline Installation for Air-Gapped Environments

Step 1: On an internet-connected machine:

ollama pull deepseek-coder-v2:16b
ollama export deepseek-coder-v2:16b -o deepseek-model.tar

Step 2: Transfer deepseek-model.tar to your secure environment

Step 3: Import on the target machine:

ollama import deepseek-coder-v2:16b deepseek-model.tar

Usage

Command Line (Single PDF)

python -m ai_dispute_platform.pipeline.cli --pdf report.pdf --output ./results

Batch Processing (Watch Folder)

python -m ai_dispute_platform.pipeline.cli --watch ./incoming --interval 5

Email Notifications

python -m ai_dispute_platform.pipeline.cli --pdf report.pdf --notify disputes@bank.com

Configure SMTP via environment variables:

export SMTP_SERVER="smtp.yourbank.com"
export SMTP_PORT="25"
export SMTP_USER="username"
export SMTP_PASSWORD="password"

Web Interface (Self-Hosted)

Run on your internal server:

python -m ai_dispute_platform.web --host 127.0.0.1 --port 8080

Open in your browser:

http://127.0.0.1:8080

Output Files

After processing, you will find these files in your output directory:

File	Purpose
output	Metro 2 file (upload to credit bureau)
output_report.txt	Human-readable text report
output_report.html	Browser-viewable report
output_report.pdf	Printable PDF report

For US Financial Institutions

Security & Deployment

This platform is designed for on-premises deployment within your secure environment:

• ✅ All processing happens behind your firewall
• ✅ No customer data ever leaves your infrastructure
• ✅ Complete audit trail of all actions
• ✅ Supports air-gapped environments

Data Residency

Your data stays on your servers. We never receive or store your customer information.

Pricing

Plan	Price	Includes
Community	$0	5 disputes/month (testing only)
Professional	$299/month	100 disputes + AI analysis
Enterprise	Custom	500+ disputes + priority support

Volume pricing: Additional disputes beyond plan are $5–$15 each depending on volume.

Contact: contact@clearwatercodes.com for custom quotes.

Support

• Documentation: This README
• Email Support: contact@clearwatercodes.com
• Paid Support: Screen-sharing guidance for installation (quote based on complexity)

License

MIT License with commercial option for volume >100 disputes/month.