Philipp Krenn

@xeraa
633 Followers
174 Following
1.7K Posts
🎩 of DevRel & Developer 🥑 at Elastic — toots about Elasticsearch, Kibana, search, observability, security
websitehttps://xeraa.net
twitterhttps://twitter.com/xeraa
linkedinhttps://www.linkedin.com/in/philippkrenn/
githubhttps://github.com/xeraa

more cursor skills in the wild — here and everywhere else. spin up an elasticsearch cluster, manage kibana, configure security, set up OTel,...
and almost hidden under it is also the brand new MCP endpoint for the elastic docs

https://cursor.com/marketplace/elastic

BM25: "why won't you die?!"
> the lexical retriever BM25 with appropriate setup outperforms neural rankers in most cases; notably, gpt-oss-20b with BM25 on the passage corpus achieves the highest answer accuracy across all retrieval settings in our study; BERT-based learned sparse and multi-vector dense retrievers generalise better than LLM-based single-vector dense retrievers; and re-ranking remains highly effective

BM25 just keeps hanging on. more in this new paper: https://arxiv.org/pdf/2602.21456

discussion of the day: communication within an AWS region on public IPs (no VPC) is "over the (public) internet"?
no: the physical network is fully controlled by AWS and traffic never leaves it
yes: the traffic uses public IPs and is globally routable, so logically it's the internet
no
0%
yes
100%
other
0%
Poll ended at .

is it an API key for a cloud account or the cluster?
way too common of a confusion. so why not both?! released today for serverless projects and your elastic cloud account

PS: please be extra careful in your day to day use with these. they have an extra wide blast radius 💥

someone woke up at RSA and chose violence: https://vibecoded.vc/cooked/ 😂
though it is entertaining (and I won't agree on all of the takes ;) )

obligatory GTC keynote tweet when you make the top slides: building HNSW graphs with up to 12x the throughput and 7x faster merges on elasticsearch and NVIDIA cuVS (a GPU-accelerated library for vector search)
powered by CAGRA (graph-based ANN algorithm built to run natively on GPUs) that still works with CPUs for search

full post: https://www.elastic.co/blog/elastic-nvidia-cuvs-integration
and find us at GTC — we have a booth and I'll be around today and tomorrow

PS: yeah, this is a lot of acronyms

nvidia GTC where every second word (spoken and on booths) is AI, agent, or token 😬

PS: who else is around?

BM25 for "sparse visual-word activations"? https://arxiv.org/abs/2603.05781
BM25 just refuses to disappear or even take the back set for retrieval
soon...
and there's already the .md version for every elastic docs page and you can download the full docs as MD in a ZIP file too
the irony on the vibecoding subreddit