Interesting findings and visualization while predicting new movies and actor collaborations using graph machine learning (#stellargraph, #pytorch geometric and #dgl). Increasingly easy to do these things (including the GDS lib in #Neo4j). #graphneuralnetworks #knowledgegraphs

Leaving X/Twitter. Been fun, but not a good place anymore.

Impressed by Kestra, the open source declarative data orchestration platform. Not yet replacing Airflow or Dagster but it will. Even has Neo4j connector out of the box. #workflow #ETL

PromptWright: generate large synthetic datasets using a local LLM (for fine tuning or other purposes). #LLMs

You can now run Llama 3.2 Vision locally via Ollama. At some point voice, video and images will all be possible, it won't take long.

NuExtract 1.5 is a multilingual content extraction model, can handle arbitrarily long documents, and outperforms GPT-4o in English while being 500 times smaller.

Wonderful network data science course taught at Johns Hopkins University. Uses Graspologic extensively, some advanced topics like graph embeddings, graph matching and lots of examples. #graphs.

CrewAI is late to the party after LangGraph and LlamaIndex flows, but now has a workflow API as well. It's based on attributes and fun to use. #LLMs

d'AIgram by Synergy Codes is an AI-powered tool that turns hand-drawn sketches, pictures, and screenshots of your flowcharts, graphs, and diagrams into editable diagrams. #diagramming.

Microsoft finally found a way to inject ads. Inevitably.

Microsoft's graph RAG project stirred a bit but it's a costly approach to context-aware RAG. The LightRAG approach discussed in this paper is metrically better than graph RAG and tons cheaper. #LLMs .

The multi-modal Unum store seems on paper a killer. The NetworkX integration is, like Memgraph, of particular interest with bindings for Python and Go. Anyone has evaluated it, what gives? @unum_cloud #GraphDatabase

Agents as intelligent message transformers in the Langroid framework. More promising than OpenAI's Swarm effort. But every week there is a new agent framework, don't put all your eggs in any of them. #LLMs

AgentLite: a lightweight tool for LLM agent systems by SalesForce AI. Simplify agent creation with easy orchestration and flexible implementation. Explore comprehensive examples to kickstart your development. #AI #LLMs

Instructor: structured outputs for #LLMs. Such as bliss. Really the glue between AI and classic programming.

Ell is an innovative approach to programming treating prompts as code and integrating LLMs as just another functionality. Such a great idea. #LLMs

SurrealDB 2.0 is out and now supports GPU inference with support for Torch, TensorFlow and SKLearn models. It's multimodal including graphs. Heaps of nice to have goodies and open source. Big fan. @SurrealDB #GraphDatabase

Alibaba continues to invest in graph analytics and machine learning on graph via its GraphScope stack. #graphs

GPU-based graph layout outperforms CPU significantly, handling hundreds of thousands of nodes and edges in mere seconds. While some may argue that force-directed layout isn't top-tier, the benefits of using GPU's are undeniable. Discover more: #Graphs

Perplexica is an AI-powered search engine. It is an open source alternative to Perplexity AI and works like a charm. If you use it with Ollama and Llama3.1 you have a fully local no-cost search engine.

Ontodia is an open source data #diagraming library designed to visualize RDF data sets (ontologies), but could be tailored to almost any data source by implementing a data provider interface. #RDF