Description

• RAG over many documents is much harder than RAG over a single document. Create a Github repo showcasing how you’re able to create a RAG pipeline over ≥ 1000 PDFs that works well!
• Document your techniques. If you do this well this will be a good reference repo for others to follow for more advanced RAG use cases! We will be happy to feature this as a LlamaPack, blog post, YouTube video, social media, or through other formats.

Acceptance Criteria

• Must use LlamaIndex (and optionally LlamaHub).
• Bonus: Contains an evaluation metric + dataset showing how this is better than other techniques.

Technical Details

• LlamaIndex contains a wide set of core abstractions and techniques to choose from for different advanced RAG use cases. Some examples include auto-retrieval, our sub-question query engine, multi-document agents. Ideally you’re able to create your own custom modules and adapt some existing techniques.
• Creating custom modules guide: https://docs.llamaindex.ai/en/latest/optimizing/custom_modules.html
• Sample production RAG guide in the docs: https://blog.llamaindex.ai/a-cheat-sheet-and-some-recipes-for-building-advanced-rag-803a9d94c41b

Prizes

In addition to the Replit Cycles, you will get:

• A Limited Edition LlamaIndex Jacket (seriously limited edition)
• Featured on socials (we have ≥ 70k followers on LinkedIn, ≥ 50k followers on Twitter)