
engineering @elastic. working on @searchkit. Github https://teapawt.org/t.co/cyMFUPZHNH
London UK
Joined November 2008
- Tweets599
- Following236
- Followers148
- Likes624
Pinned Tweet
Searchkit: React Components for elasticsearch. Make fast usable searches quick! searchkit.co
Joe McElroy retweeted
Domain-specific generative AI: comparing pre-training, fine-tuning, and RAG as approaches for storing and retrieving knowledge from LLMs. Learn more: go.es.io/47EQJn6
#ElasticSearchLabs #GenerativeAI
Joe McElroy retweeted
Vector Similarity Computations FMA-style elastic.co/search-labs/blog/…
Joe McElroy retweeted
🦜🏁Self Query Template (@elastic)
Often times documents come with really useful metadata! But how can you best take advantage of this metadata when doing RAG? 🤔
Self-query is one way to do so. This converts a 🗨️natural language query into one piece that is embedded and compared via cosine similarity, just as in naive RAG. But then a "metadata filter" is also extracted, that can be applied to metadata present in the vectorstore
❓This is useful because often times a query may not only refer to semantic meaning, but could also reference some of the metadata
For example: "what is a movie about aliens from the year 1980" 👽
While "aliens" may be a semantic query, "year == 1980" is not, this is an explicit filter that should be applied
Most vectorstores allow for some metadata filters, so this can be used with most of them
Check out our recent template which does this for ElasticSearch: github.com/langchain-ai/lang…
This is an example "query construction" - check out @RLanceMartin's awesome post on this from earlier in the week: blog.langchain.dev/query-con…
Joe McElroy retweeted
Interested in multilingual vector search with #Elasticsearch? New blog post drop featuring @Microsoft's multilingual E5 embedding model. Have a read, run the notebook, und suche in deiner Sprache! 😉
search-labs.elastic.co/searc…
Joe McElroy retweeted
LlamaIndex + @elastic 🚀
This has been a highly requested feature (a lot of our users are using Elasticsearch!)
HUGE shoutout to @phoey1 for making this happen.
- Vector/text/hybrid search
- Metadata + metadata filtering
- es_filters
- And more!
gpt-index.readthedocs.io/en/…
Joe McElroy retweeted
The latest and greatest in 🦜🔗:
🧠 @neural_internet Bittensor LLM
⚡ DashVector store
🤖 @zep_ai vector store
♻️ @elastic search self-querying retriever
a 🧵:
Joe McElroy retweeted
♻️ @elastic search self-querying retriever
Combine the power of the Elasticsearch vector store with LangChain’s query constructor chain to create a self-querying retriever.
Huge s/o to @phoey1 for continuing to add Elasticsearch features to 🦜🔗!
Docs: python.langchain.com/docs/mo…
Joe McElroy retweeted
more #elasticsearch integrations coming your way :)
PS: we'll also have LangChain at our upcoming ElasticON AI event in san francisco: elastic.co/events/elasticon-…
Replying to @LangChain
🤹 New @elastic vector store
@phoey1 has done a huge revamp and added a new Elasticsearch-backed vector store, allowing for a number of new index / query strategies.
Try it out:
``from langchain.vectorstores import ElasticsearchStore``
Docs: python.langchain.com/docs/in…
Joe McElroy retweeted
New in 🦜🔗==0.0.265!
3 new model wrappers and vector store improvements:
🐾 Ernie chat model
🚅 @berriai LiteLLM chat model support
🪄 @neuralmagic DeepSparse LLM
🤹 Major @elastic vector store improvements
a 🧵:
Joe McElroy retweeted
🤹 New @elastic vector store
@phoey1 has done a huge revamp and added a new Elasticsearch-backed vector store, allowing for a number of new index / query strategies.
Try it out:
``from langchain.vectorstores import ElasticsearchStore``
Docs: python.langchain.com/docs/in…
Joe McElroy retweeted
This is why optimization tasks are addictive! Getting some nice improvements for vector search in Apache Lucene 9.7. github.com/erikbern/ann-benc…
Joe McElroy retweeted
Why we designed, in the last couple of years, Elasticsearch and Lucene vector database implementation the way we did, and where we are heading elastic.co/fr/blog/vector-se…
Joe McElroy retweeted
I ❤️Lucene.
Like so many times before (real time search, doc values, bm25, bkd, compressions, …), we are actively investing in making Apache Lucene also a great Vector Database elastic.co/blog/accelerating…
Joe McElroy retweeted
If I don’t say it enough, this is a fantastic resource lnkd.in/gX6c8fqA
Joe McElroy retweeted
Asking a question and getting back a list of search results already seems so old-fashioned.
Joe McElroy retweeted
It’s been a jam-packed week at #MSBuild! We launched the Elasticsearch Relevance Engine (ESRE) and our team has been busy connecting with attendees to share how ESRE can help build better #AI powered search applications. Learn more: go.es.io/43q7Ffi
Joe McElroy retweeted
Search has never been more relevant in a world of generative AI. Very excited to announce Elasticsearch Relevant Engine, bundling a lot of the work we have done over the past couple of years, and all the exciting work to come
Say hey 👋 to ESRE (sounds like ez-ray) — the Elasticsearch Relevance Engine. The product of more than two years of research and development, it gives developers the tools they need to build AI search applications that integrate with GAI.
Learn more → go.es.io/3BTBKbj
Introducing ESRE — the Elasticsearch Relevance Engine
Joe McElroy retweeted
This will be interesting trec-product-search.github.i…























