Comparing Three AI-Powered Solutions for Car Manual Queries 🤖🚗

Welcome to the Comparing Three AI-Powered Solutions for Car Manual Queries project! This repository contains the codes and resources that have been used and adapted for the analysis and comparison of three leading AI-based solutions for answering technical questions related to car manuals in PDF format. Our goal is to help you make an informed decision about the best solution for your specific needs.

🤖 Solutions Overview

We've evaluated the following three AI-powered solutions:

1. Doc Chatbot with LlamaIndex: A chatbot that uses LlamaIndex and LangChain libraries to connect large language models (LLMs) to external data sources like PDF car manuals, enabling the creation of chatbots that can learn from diverse data sources and deliver good answers. The code for this solution can be found in the doc-chatbot-with-llamaindex folder.

2. Ask your PDF: A graphical interface that combines Streamlit and Langchain to allow users to upload a PDF file and ask questions about its content. The application extracts text from the PDF, creates embeddings using the OpenAIEmbeddings library, and stores them in a FAISS database for similarity search. The code for this solution can be found in the ask-your-pdf folder.

3. Question and answer system: A system that combines LangChain and Large Language Models, such as OpenAI’s GPT-3, and uses React/Typescript for the frontend and Python FastAPI for the backend. QDrant is used to store embeddings and text documents for fast search. The code for this solution can be found in the QA-system folder.

For detailed documentation and resources for each solution, please refer to the README.md files inside each solution's folder. You will also find the 'How to Run' instructions within each solution's README.md file.

📊 Analysis and Comparison

Our analysis and comparison of the three solutions focused on the following criteria:

1. Performance
2. Cost
3. Answer Accuracy
4. User Experience

Criteria	Doc Chatbot with LlamaIndex	Ask your PDF	Question and answer system
Performance	Good	Good	Good
Cost	$$$	$$	$
Answer Accuracy	High	High	Medium
User Experience	Fair	Good	Good

You can find a detailed report of our analysis here.

✉️ Contact Information

If you have any questions or feedback, please feel free to reach out to the authors of this analysis:

• Mariana Azevedo
• Morsinaldo Medeiros
• Thaís Medeiros

The following are the links to the sources where the codes were taken from:

• Doc Chatbot with LlamaIndex
• Ask your PDF
• Question and answer system