RAG-based Knowledge Extraction Challenge 🚀

Context and Motivation 🌟

Welcome to the SelectCode RAG-based Knowledge Extraction Challenge! 🎉 At SelectCode, we're facing a common corporate challenge: making technical knowledge more accessbible. We're turning to you, our innovative applicants, to help us address this using Retrieval Augmented Generation (RAG) techniques.

Challenge Description 📋

Your task is to design and implement a data pipeline that processes technical documentation and uses RAG to let people ask questions on how to perform some of these technical tasks.

Technical Requirements 🛠️

• You will be provided with multiple guides in markdown format (realistic, imperfect output from a PDF extractor). You can find them here (opens in a new tab).
  • You do not have to use all of these files. You can just choose the ones you like best.
• You have the freedom to choose your preferred programming languages, frameworks, and tools.
  • Focus on using RAG/LLM tools that are open-source or provide free API keys for evaluation, using paid service that might provide better output will not be an advantage.
• Your solution should demonstrate a well-designed data pipeline that incorporates RAG techniques.
  • You should pay attention to the classic RAG workflow like preprocessing the data if needed, generating embeddings, ingesting chunks into a vector database, etc.
  • Optional techniques that show domain knowledge are also appreciated (e.g. reranking of results)

Core Functionality ✅

1. File Processing: Ability to ingest and process markdown guides.
2. RAG Implementation: Utilize RAG techniques to analyze and generate insights from the guides.
3. Output Generation: Produce some form of output that answers technical questions about the contents of the guides (e.g. how do I use branches in git?).

Implementation Guidelines 📝

• Focus on creating a robust and scalable data pipeline.
• Your code should be of production quality. Pay attention to:
  • Clean, readable, and well-structured code organization
  • Appropriate commenting and documentation
  • Error handling and logging
• Consider best practices for data processing and RAG implementation.
• Be creative! We're interested in novel approaches to making meetings more efficient or enjoyable.

Evaluation Criteria 🏆

We will evaluate your work based on the following:

• Functional quality: How well does your solution address the problem of extracting and accessing knowledge?
• Code quality: We're looking for clean, readable, and well-structured code.
• Pipeline design: How well-designed and scalable is your data processing pipeline?
• RAG implementation: How effectively have you utilized RAG techniques?
• Documentation quality: We're looking for a well-structured README.md file that explains your approach and how to use your solution.
• Problem solving: We want to see your thought process, so commit early and often with clear, informative commit messages.

Submission 📤

• Provide your code in a public GitHub / GitLab repository.
• Include a README file that contains:
  • An overview of your solution's architecture
  • Installation and usage instructions
  • Explanation of your RAG implementation and how it addresses the challenge
  • Any assumptions you made and why
  • Ideas for future improvements or extensions of your solution
• If you've implemented any particularly unique or innovative features, please highlight them in your README. 🌟
• Send all this information via email to Stephan ([email protected]).

Questions? ❓

If you have any questions or need clarification, please don't hesitate to reach out. We're here to help! 📧

Good luck with the challenge! We're excited to see your creative and effective solutions to improve knowledge access. 🍀🚀