ask-my-pdf

Ask my PDF: A Detailed Overview

Introduction

Ask my PDF is a unique application designed as a Proof of Concept system aimed at enhancing users' experience with finding answers to questions about board game rules. Though the application remains a work-in-progress and may contain certain bugs or incomplete features, its creator invites feedback and follows on Twitter for updates and news.

Purpose and Functionality

The primary purpose of Ask my PDF is to assist users in interpreting and answering questions related to board game rules using the game's instruction manuals. This focus leverages the developer’s passion for board games, ensuring that the application serves a dedicated niche community. Despite the potential limitations of the model, such as hallucinations, this use case remains relatively benign and straightforward to manage.

Users can access the application on the Streamlit Community Cloud at https://ask-my-pdf.streamlit.app/, where it functions with the support of OpenAI’s GPT-3. To utilize the app, users must have a valid OpenAI's API key.

Underlying Technologies

Ask my PDF incorporates cutting-edge technologies and techniques outlined in several academic papers:

• In-Context Retrieval-Augmented Language Models (RALM): Detailed in this paper, RALM is a technique facilitating sophisticated retrieval-augmented interactions, crucial for providing accurate responses.

• Hypothetical Document Embeddings (HyDE): This method, detailed in another paper, involves a precise zero-shot dense retrieval methodology that does not rely on relevance labels, providing a robust framework for handling queries efficiently.

How to Install and Run

To install and get Ask my PDF running on a local system, follow these steps:

1. Clone the repository using:

   git clone https://github.com/mobarski/ask-my-pdf
   

2. Install the required dependencies:

   pip install -r ask-my-pdf/requirements.txt
   

3. Start the application:
   • Navigate to the source directory:

     cd ask-my-pdf/src
     

   • Run the app using either:

     run.sh
     

     or

     run.bat
     

High-Level Documentation

RALM + HyDE

This integration facilitates enhanced interaction by combining RALM with HyDE, depicted in the accompanying diagrams provided in the app’s documentation.

RALM + HyDE + Context

Adding context to the RALM and HyDE combination allows for even richer interaction dynamics, illustrated as follows:

Configuration and Environment Variables

Ask my PDF relies on several environment variables for configuration, divided into categories based on generalized and specific configurations like local filesystem or cloud-based storages.

General Configuration:

• STORAGE_SALT: Used for deriving cryptographic salts.
• STORAGE_MODE: Specifies the storage mode which could be S3, LOCAL, or DICT.
• STATS_MODE / FEEDBACK_MODE / CACHE_MODE: Various modes to manage statistics, feedback, and caching using REDIS or DICT.

Local Filesystem and S3 Configuration:

• Involves setting paths for storage and cache, along with detailed S3 configuration for cloud storage.

Redis Configuration:

• Involves specifying the Redis database URL for persistent data handling like usage statistics or user feedback.

Community Version Options:

• OPENAI_KEY, COMMUNITY_DAILY_USD, and COMMUNITY_USER: These variables configure the default settings for community users accessing the application.

Conclusion

Ask my PDF stands as a promising solution for board game enthusiasts seeking support with rule interpretations. Its integration of advanced language processing models and flexible setup options demonstrates the potential for further development and wider application beyond its initial focus.