Generative AI Augmented Retrieval - GAR presi

Attached are slides+notebook from my talk “Introducing GAR (Generative AI Augmented Retrieval)” at the Systematically Improving RAG Applications course.

GenerativeAI Augmented Retrieval uses Generative AI to make boring-old search better.

Enjoy!

Slides

Notebooks

I encourage you to play with the data here and share your learnings. You’re libel to teach me a thing or two!

Baseline data from the WANDS (Wayfair Furniture Search) dataset.

• WANDS Baseline

Didn’t work

Shooting in the dark with random ideas usually doesn’t work. You need to look at your data.

• Naive content generation
• Naive synonym generation.

Actually works

(Works becuse I looked at teh data and solved actual problems present in the BM25 baseline :) )

• Cleanup title (LLM generates new title)
• Use Google Product Taxonomy