Neo4j GenAI Workshop

Please see genai-workshop.ipynb which serves as the self-contained workshop.

The other companion notebooks contain code for staging data, building the Neo4j Graph, and providing easy access to demos:

data-prep.ipynb stages the workshop data, sampling and formatting data sourced from the H&M Personalized Fashion Recommendations Dataset.
data-load.ipynb loads the staged data into Neo4j, performs text embedding, and creates a vector index.
genai-workshop-w-outputs.ipynb is a reference notebook for following along and checking outputs. It is simply genai-workshop.ipynbwith cell outputs intact.
genai-example-app-only.ipynb is a copy of genai-workshop.ipynb that contains only the final section: the demo LLM GraphRAG app for content generation. It assumes you have already run genai-workshop.ipynb and exists only for instructor demo purposes.

Changelog

Transition from using Neo4j Sandbox to AuraDS
Split out data loading
- Split out data load into a separate notebook
- Live workshops now begin with the dataset pre-loaded to cut down on time and spend more of the course walking through GraphRAG. the data-load.ipynb is kept for reference and replication.
- Remove neo4j_tools Python package. the functions/utilities are now included in data-load.ipynb
Updated Workshop Slides
Added more query exploration & improved explainer queries
- Add Browser-based graph exploration in beginning of workshop
- Include database tips & more Cypher queries in multiple steps
- Update explainer markdown and code cells for graph patterns and GDS for clarity
- Various other minor adjustments to markdown and code to improve course quality
Added genai-workshop-w-outputs.ipynb and GitHub Actions Workflow
- genai-workshp.ipynb is now maintained with cleared outputs for better workshop experience and easier PR review
- A GitHub actions workflow automatically tests data loading, workshop, and example-only notebooks and auto-commits genai-workshop-w-outputs.ipynb file for each PR.

improve LLM response quality and cleaned up code for LLM chains and vector stores
- parameterizing customer id so don't need to recreate chains & stores for each customer
- updated prompts to better account for seasonality and use all retrieved data
- update to use gpt-4o
Improve text embedding speed and reduce code by transitioning to native genai.vector Cypher functions
Updated slides
Various other minor adjustments to markdown and code to improve course quality

(fix) Add langchain_community to the libraries that are pip installed in the notebooks
Simplify and Shorten Course
- Shortened GDS section to just three cells to run
- Condensed Vector Search Section
- Condensed Loading to Single Notebook Cell
- Switched Recommendation Retriever to a Simple KG Query
- Adding neo4j_tools Package to hold convenience functions for loading data and reduce code footprint in main workshop notebook
- Updated to GPT-4 throughout
- General Notebook Cleaning - Removed duplicate load statements, updating to newest llm packages, etc.
Provide Better Explainers & Examples
- Add A Chain for Printing Final Prompt to LLM with retrieval data to better explain process.
- Added Differentiated Names for Customer Examples in Demo App.
Added Additional Resources
- Added workshop slides
- Added "demo only" notebook

Contributions are welcome! To contribute please:

Make a PR with a descriptive name
If you are updating genai-workshop.ipynb please ensure to clear all outputs before committing.
Do not alter the genai-workshop-w-outputs.ipynb file. This file is autogenerated upon creating/updating PRs.