Top Tweets for #Strwythura

9 months ago

We love seeing yFiles for #Streamlit featured in the #Strwythura project to provide visual oversight for entity-resolved #KnowledgeGraphs. 👉 Try the free yFiles integration for Streamlit here: https://t.co/LYqKQPk6al #GraphAI #RAG #yFiles #DataEngineering #StreamlitApp

Connected Data

@Connected_Data

9 months ago

Strwythura: build a Streamlit app for a question/answer chatbot about a specific topic using advanced techniques for knowledge graph and embeddings Many tutorials about GraphRAG recommend delegating the construction of knowledge graph to large language models (LLM). This seems backwards, since you probably already have structured data and other context about your domain on hand. Why discard useful information, then acquiesce to whatever an LLM hallucinates instead? You’re spending more money and time, but getting less accurate results in return. That approach makes no sense in terms of robust engineering. Instead, this tutorial by @pacoid unbundles the processes needed to reach a working AI application. It explains how to construct an entity resolved knowledge graph from structured data sources and unstructured content sources, implementing an ontology pipeline, plus context engineering for optimizing AI application outcomes within a specific domain. The process is enriched by using entity embeddings and graph algorithms to develop an enhanced GraphRAG approach, which implements a question/answer chatbot about a particular domain. This material provides hands-on experience with advanced techniques as well as working code you can use elsewhere. * Using sophisticated NLP pipelines based on spaCy, GLiNER, textgraphs, and related libraries when extracting information from unstructured content. * Leveraging computable semantics based on standards such as RDF, SKOS, etc., to develop and apply domain-specific semantics which guide the embedding models and language models. * Applying graph analytics to make inference from the knowledge graph, to augment the vector embeddings results before presenting content to an LLM for question/answer summarization. * Using declarative methods for LLM integrations (for example, based on DSPy) instead of spending loads of time on exotic prompt definitions. Then using an observability framework (for example, based on Opik) to collect evaluations and experiments, establishing a feedback loop for optimizing the prompts and weights used in LLM integrations. These techniques provide results which are provably better/faster/cheaper than following an “LLM-everything” approach, plus much more oversight for the intentional arrangement of a knowledge graph. This runs locally without lots of cost, including MLOps instrumentation. The code can be easily extended for other AI app use cases, other topics, and more integrations. The workflow begins with entity resolution, then goes on to illustrate lots of useful graph technologies in practice, combined with NetworkX, spaCy, GLiNER, DSPy, LanceDB, Opik, yWorks, etc. See also - Combining Data from Structured and Unstructured Sources to create High-Quality Knowledge Graphs https://t.co/xICpEpqnsB #GraphRAG #LLMs #Python #NLP #MLOps #DataEngineering #MachineLearning -- Connected Data London 2025 brought together leaders and innovators. Were you there? 🎥 Watch the sessions: https://t.co/w61JGmLcxh 📩 Join the community: https://t.co/RMj1EA5vl1 Join community legends and new voices in #CDL25 for all things #KnowledgeGraph #Graph #analytics #datascience #AI #graphDB #SemTech #Ontology

Connected_Data's tweet photo. Strwythura: build a Streamlit app for a question/answer chatbot about a specific topic using advanced techniques for knowledge graph and embeddings

Many tutorials about GraphRAG recommend delegating the construction of knowledge graph to large language models (LLM). This seems backwards, since you probably already have structured data and other context about your domain on hand.

Why discard useful information, then acquiesce to whatever an LLM hallucinates instead? You’re spending more money and time, but getting less accurate results in return. That approach makes no sense in terms of robust engineering.

Instead, this tutorial by @pacoid unbundles the processes needed to reach a working AI application. It explains how to construct an entity resolved knowledge graph from structured data sources and unstructured content sources, implementing an ontology pipeline, plus context engineering for optimizing AI application outcomes within a specific domain.

The process is enriched by using entity embeddings and graph algorithms to develop an enhanced GraphRAG approach, which implements a question/answer chatbot about a particular domain. This material provides hands-on experience with advanced techniques as well as working code you can use elsewhere.

* Using sophisticated NLP pipelines based on spaCy, GLiNER, textgraphs, and related libraries when extracting information from unstructured content.
* Leveraging computable semantics based on standards such as RDF, SKOS, etc., to develop and apply domain-specific semantics which guide the embedding models and language models.
* Applying graph analytics to make inference from the knowledge graph, to augment the vector embeddings results before presenting content to an LLM for question/answer summarization.
* Using declarative methods for LLM integrations (for example, based on DSPy) instead of spending loads of time on exotic prompt definitions. Then using an observability framework (for example, based on Opik) to collect evaluations and experiments, establishing a feedback loop for optimizing the prompts and weights used in LLM integrations.

These techniques provide results which are provably better/faster/cheaper than following an “LLM-everything” approach, plus much more oversight for the intentional arrangement of a knowledge graph.

This runs locally without lots of cost, including MLOps instrumentation. The code can be easily extended for other AI app use cases, other topics, and more integrations.

The workflow begins with entity resolution, then goes on to illustrate lots of useful graph technologies in practice, combined with NetworkX, spaCy, GLiNER, DSPy, LanceDB, Opik, yWorks, etc.

See also - Combining Data from Structured and Unstructured Sources to create High-Quality Knowledge Graphs

https://t.co/xICpEpqnsB

#GraphRAG #LLMs #Python #NLP #MLOps #DataEngineering #MachineLearning

--

Connected Data London 2025 brought together leaders and innovators. Were you there?

🎥 Watch the sessions: https://t.co/w61JGmLcxh
📩 Join the community: https://t.co/RMj1EA5vl1

Join community legends and new voices in #CDL25 for all things #KnowledgeGraph #Graph #analytics #datascience #AI #graphDB #SemTech #Ontology

504

Top Tweets for #Strwythura

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users