Hypermode Live: Knowledge Graphs + AI

Video Summary

The video "Hypermode Live: Knowledge Graphs + AI" features discussions on the integration of knowledge graphs with AI, particularly focusing on a hackathon launched by Hypermode and Neo4j.

Key Concepts and Features Discussed:

• Knowledge Graphs and AI Integration:
  • Knowledge graphs can complement large language models (LLMs) by providing memory and contextual information that LLMs often lack.
  • The combination of graphs and LLMs can enhance the performance of AI applications.
• Hackathon Announcement:
  • A hackathon focused on building model-native applications using knowledge graphs and AI models has been launched, with a prize pool of $10,000.
  • Participants are encouraged to use the open-source framework "Modus" and can leverage tools such as Neo4j and Dgraph.
• Modus Framework:
  • Modus is a serverless framework designed for building model-native applications.
  • It supports multiple programming languages and allows users to create functions that interact with various data sources, including knowledge graphs.
  • The framework utilizes WebAssembly for efficient function execution.
• Hands-On Examples:
  • Michael demonstrated how to use Neo4j with Modus to generate embeddings from movie plots and perform vector searches to find similar movies.
  • Raphael showcased a project using Dgraph to manage markdown documents, extracting relevant information and generating responses based on user queries.
• Best Practices for Document Chunking:
  • When preparing documents for vectorization, it is crucial to maintain semantic coherence by using natural document structures (e.g., paragraphs, sections) as boundaries for chunks.
  • Different chunking strategies can be employed, including fixed-size chunking, recursive chunking, and semantic chunking based on vector similarity.

Resources and Guides:

• The video includes links to the Modus GitHub repository for code samples and further resources.
• Participants in the hackathon can find starter code and guidelines on the Devpost platform.

Main Speakers:

• Will (Hypermode)
• Michael Hunger (Neo4j)
• Raphael (Hypermode)

This session serves as both an informative discussion on the potential of knowledge graphs in AI and a practical guide for developers looking to participate in the hackathon.