🧠 LLM-Powered Email Extraction App

This Streamlit web app lets you upload .eml email files and use Anthropic Claude (via AWS Bedrock) to:

• Extract structured JSON data from emails
• Automatically generate cleaner, reusable prompts
• Compare prompt quality side-by-side in a 4-column interface

---

🚀 Features

• ✅ .eml File Upload & Parsing — Extracts text, html, sender, subject, attachments (PDFs included)
• 🤖 Claude-Powered Extraction — Uses user_prompt to pull out JSON info from emails
• ✨ Improved Prompt Generator — Claude rewrites your prompt in scalable, few-shot natural language
• 📊 Prompt Comparison Tool — See a side-by-side markdown table comparing your prompt vs Claude’s
• 🖼 Fixed 4-Column UI — Horizontal layout: Email Preview | JSON | Improved Prompt | Comparison
• 🌙 Dark Panel Layout — Panels styled for readability in dark mode, no HTML clutter

---

🛠️ Technologies Used

Tool	Purpose
Streamlit	Frontend Web App
AWS Bedrock	Claude 3 Haiku (LLM prompt processing)
boto3	Python SDK for AWS
PyPDF2	PDF Attachment Parsing
BeautifulSoup	HTML to Text Conversion
Bleach	HTML Sanitization (optional)

---

📦 Installation

🐍 Python Environment

git clone https://github.com/rsonic17/llm-prompt-testing-utility.git
cd llm-prompt-testing-utility
python -m venv venv
source venv/bin/activate   # or venv\Scripts\activate on Windows
pip install -r requirements.txt

🔐 Setup AWS Credentials

Ensure you have:

• AWS credentials in ~/.aws/credentials
• AWS region and model ID set via environment:

export AWS_REGION=us-east-1
export BEDROCK_MODEL_ID=anthropic.claude-3-haiku-20240307-v1:0

You can also use a .env file and python-dotenv.

---

🧪 Usage

streamlit run app.py

1. Upload a .eml file
2. Paste a Claude-compatible prompt (use {email_data} as placeholder)
3. Click:
   • 🧠 Extract with LLM
   • ✨ Suggest Better Prompt
   • 📝 Compare Prompts
4. View results in 4 scrollable panels

---

🔤 Prompt Format (Example)

Paste this in the "User Prompt" box:

Extract the following fields from this email:
- Buyer Name
- Sender Email
- Date
- Payment Amount

Output JSON like:

{
  "buyer": "...",
  "sender": "...",
  "amount": "...",
  "date": "..."
}

Email:
{email_data}

---

🖼 UI Overview

📸 You can add screenshots here later.

Panel	Purpose
📄 Email Preview	Raw email text
📦 LLM Extracted Data	Claude’s structured JSON
🌟 Improved Prompt	Claude's few-shot rewrite
📑 Prompt Comparison	Markdown table comparison

---

👥 Contributing

Pull requests are welcome. Please open an issue to discuss any major changes.

---

📄 License

MIT License © 2025 [Your Name / Org]