Nexteer Capstone Project - Leveraging AI/ChatGPT

Overview

This project automates the analysis of procurement data from PowerBI dashboards using AI and Large Language Models. It significantly improves efficiency in generating insightful summaries from complex data, thus enhancing operational efficiency in report generation. The project was completed for Nexteer Automotive for the Carnegie Mellon University Master of Information Systems Management capstone project.

Authors

• Charles DeVries
• Evelyn Sun
• Kathy Wang
• Muhammad Asad Shoaib
• Sujai Sheel Chaudhary

Getting Started

Prerequisites

• Python 3.x
• Pip package manager

Installation

1. Extract the Zip File

   • Extract the contents of the provided zip file to a desired location on your computer.

2. Install Required Libraries

   • Navigate to the extracted project directory and install the required libraries using the requirements.txt file.

   cd PATH_TO_PROJECT_DIRECTORY
   pip install -r requirements.txt

3. Acquiring AskMyPDF API Key

   • This application uses the AskMyPDF API for processing PDF documents. You need to obtain an API key from AskMyPDF.
   • Visit AskMyPDF Pricing and subscribe to the $20 per month plan.
   • After completing the sign-up and payment, you will receive an API key.
   • Insert this API key into ui.py where indicated (api_key = "...").
   • The current API key is set to expire in December

4. Preparing Data

   • To use the dashboard, populate the "kpis" folder with PDF exports from the Global KPI Dashboard.
   • Within PowerBI, export the pages with "(LLM View)" in the name.
   • Use "Export" > "PDF" in PowerBI, ensuring "Exclude hidden report tabs" and "Only export current page" are checked.
   • Rename these exported pages according to the following mapping and add them to the KPI folder:
     • Sourcing On Time -> sourcing.pdf
     • Perfect Launch - PPAP On Time Summary -> ppap.pdf
     • Gate Review On Time -> gate.pdf
     • Perfect Launch - Meet Costbook -> meet.pdf
     • Overall -> overall.pdf
   • Failure to add or correctly name the files will result in an error.

5. Running the Application

   • Run the Streamlit application from the /repo directory using the following command:

   streamlit run ui.py

6. Accessing the Application

   • Streamlit will start a local server. Access the application in a web browser at http://localhost:8501.

Usage

• Follow on-screen instructions in the Streamlit application to generate summaries from the PowerBI dashboards.
• To generate the final executive summary, generate the individual summaries for each KPI first.
• To edit the prompting for any summary, use the "Edit prompts?" checkbox at the bottom of the dashboard.

Disclaimer

• Information generated by Large Language Models can be prone to inaccuracies or hallucinations. While measures have been taken to reduce this risk, it is not entirely eliminated.
• If the summary contains numbers not found in the source text of the PDF, a warning will be displayed, identifying which numbers could not be matched.
• While this mechanism can reduce the risk posed by hallucinations, it does not guarantee accuracy. Ensure you verify claims made in the summary manually as well.