Joint work with Vincent Emonet, Ruijie Wang, Ana Claudia Sima, and Tarcisio Mendes at SIB Swiss Institute of Bioinformatics. #KnowledgeGraphs #SPARQL #Text2SPARQL #LLM #SemanticWeb #Bioinformatics #OpenSource [6/6]

Opsis is a one-file faceted browser for #RDF graphs. Every click is a #SPARQL query. Fully data-centric. Even its settings are part of the graph. I'm using it daily to browse my local personal knowledge graph. Now it's available for everyone to use. github.com/kvistgaard/opsis

Nodica is an open-source app that visualizes #RDF graphs with image-filled nodes. You can load the graph from a #SPARQL endpoint or from an RDF file. It's Turtle all the way down. kvistgaard.github.io/nodica/

It started as a system to help me deal with my health issues and became my OS: Retinue is a #sparql backed framework for AI agents. Learning about #RAG I asked Fable 5: github.com/Retinue-OS/retinu… how is this different from RAG? The core difference: RAG retrieves *text* for the LLM to read; this computes *answers* the LLM only interprets. In RAG, you embed chunks, do similarity search, and stuff the top-k passages into context — the LLM then synthesizes something plausible from what happened to be retrieved. That fundamentally can't answer aggregate questions: no amount of chunk retrieval gets you "you slept longer than 78% of nights," [github](github.com/Retinue-OS/retinu…) because that needs the full distribution at query time, and pasting a year of five-minute CGM readings into a prompt [github](github.com/Retinue-OS/retinu…) isn't retrieval, it's a haystack. Here the agent writes SPARQL, and the store does exact joins, GROUP BY, and counts over years of observations — the store does the aggregation, the agent does the reasoning [github](github.com/Retinue-OS/retinu…) . The result is deterministic and can't hallucinate. The second difference is the semantic layer. RAG retrieval is fuzzy over unstructured text; here everything is normalized into shared vocabularies (SOSA, LOINC, SNOMED), so a glucose reading from a CGM, a sleep score from a ring, and a step count from a watch are all the same shape and join without negotiation. Plus provenance is structural: each file's triples land in a named graph derived from the file's path, rather than chunk metadata bolted on. The doc's own summary is apt: grep gets you text, SPARQL gets you joins — and RAG is essentially fuzzy grep. They're complementary, though: this architecture deliberately leaves the prose bodies of notes unstructured, and answering questions *about prose* is exactly where RAG-style retrieval still fits. Structured facts → SPARQL; narrative text → retrieval.

Japan's @IPAjp proposes Open Data Spaces: share data across organizations without centralizing it. Every domain pairs raw data with an #RDF ontology. #SPARQL federates queries across domains at runtime. LLM agents bridge vocabulary gaps between ontologies.

🚨 Federated SPARQL queries over public SPARQL endpoints are becoming increasingly difficult to execute, which brings the fundamental motivations behind #RDF, #SPARQL, and #KnowledgeGraphs into question. In my new blog post, I explore this problem. rubensworks.net/blog/2026/04…

Querying Gmail with #SPARQL. Soon 😎

In questo articolo mostriamo una pipeline controllata per tradurre domande in linguaggio naturale in query #SPARQL eseguibili su #wikidata #fontistoriche 👉 tinyurl.com/mrz5ty59

We're happy to announce the release of Comunica version 5.0, which has full SPARQL/RDF 1.2 support, more modular query parsing, improved HTTP caching, and more! comunica.dev/blog/2026-01-07… #SPARQL #SPARQL12 #Query #JavaScript

🚀 #RDFox v7.5 is out now and it has a diverse array of incredible new additions! ✔️ #SPARQL Federation ✔️ Improved monitoring and #logging ✔️ Server-to-server #authentication ✔️ Semantic #similarity (experimental) Full release notes: hubs.li/Q03XBdRc0

Love #SPARQL and #RDF? I do. This info was super hard to find and handy. Read this. Sparqloscope: A generic benchmark for the comprehensive and concise performance evaluation of SPARQL engines Hannah Bast Johannes Kalmbach  Robin Textor-Falconi Christoph Ullinger University of Freiburg, Freiburg im Breisgau, Germany ad-publications.cs.uni-freib…

Super proud to share a new QLever milestone. We loaded one trillion triples on a single AMD Ryzen 9 machine with 16 cores and 128 GB RAM 🚀 Index size is 6.5TB and the dataset runs even with less memory. Dataset: qlever.dev/one-trillion/ #RDF #SPARQL #KnowledgeGraphs

The Weekly Edge: RDF Edition! This week’s edition is all about #RDF triplestores and more, with hits from #ISWC, #SPARQL 1.2, the @Netflix knowledge graph, the #LPG vs RDF debate, and a hot take on #OWL. Link below! 🧵

🏅 Congratulations to the winners of the #ISWC2025 SemTab Challenge! 🎓 “ADFr: Knowledge Graph Entity Linking via Interactive Reasoning and Exploration with GRASP” by Sebastian Walter and Hannah Bast. 👏 A big round of applause to the authors! #EntityLinking #SPARQL

Now finishing the second part on Façade-X, which is the 5th essay in the Containment series, I remembered that my best attempt to explain #SPARQL for non-technical users also relied on containers: strategicstructures.com/?p=1…

Here's our Github repo for viewing the Agents.md based Agent description in markdown . github.com/OpenLinkSoftware/… #SPARQL #GenAI #SmartAgent #RAG #GraphRAG #KnowledgeGraphs #LinkedData #SemanticWeb

🔥 Generative AI can now answer complex scientific questions! Our latest paper evaluates the ability of #LLMs to answer questions over #KnowledgeGraphs through Natural Language to #SPARQL translation. Excellent results across multiple benchmarks. 📄 dl.acm.org/doi/10.1145/37579…

Are object properties the introverts at the SPARQL party, waiting to be approached with a query?" Source: devhubby.com/thread/how-to-g… #MachineLearning #QueryLanguage #SPARQL #AI #properties #class

We just finished our monthly query research livestream, you can find the recording here: piped.video/live/gMQdm1XPLks Our next livestream is scheduled on September 30 at 9:30 (Brussels time): piped.video/live/DSgDZsHYgxc #SPARQL #Web #Decentralization #Query

WDBench results are in... Tentris: “I’ll just handle 1.25B triples and be 5-10x faster while I’m at it.” Everyone else: 🐢💤 We came to play. #Tentris #SPARQL #RDF #GraphDatabases #SemanticWeb