CVP & Distinguished Engineer @ Microsoft. Working on knowledge systems for AI agents. Generally curious. Opinions are my own.
Joined July 2007
- Tweets707
- Following133
- Followers1.4K
- Likes89
Let the query decide: Introducing "auto", a new retrieval reasoning effort mode in Foundry IQ that recovers 97.6% of the evidence recall of the highest effort level while using 46.9% fewer query planning tokens.
techcommunity.microsoft.com/…
Autoresearcher: great progress! 1.0 recall, 0ms latency, I just hardcoded your workload.
It if looks like ML, it has the same problems as ML, overfitting is a thing for generated code. Define your tasks and constraints carefully
Pablo Castro retweeted
We're proud to announce the availability of an end-to-end LlamaIndex + @Azure solution at #MSIgnite today!
Learn more about our integration with Azure Open AI, Azure AI Embeddings, Azure AI Search and more!
If you're at Microsoft Ignite this week @seldo will be around all week, so feel free to reach out and meet up!
llamaindex.ai/blog/announcin…
cc @msPartner
Pablo Castro retweeted
Hear from the leaders behind Microsoft's developer tools (@pmc, @yina_arenas, @jmillerdev, @toddanglin) this Thursday, November 7th at 9:00 AM PT: msft.it/6012WXeus
#Microsoft365Dev
Pablo Castro retweeted
The new GPT-4o realtime API for audio enables developers to build apps with speech-to-speech interactions. Learn how to build your own VoiceRAG app using the new model here: msft.it/6014m7BYC
ALT An image of a web interface displaying the Azure logo and a message reading 'Talk to your data' in gradient text. Below, there is a purple button with a microphone icon for voice input. The subtext instructs users to ask questions about 'Contoso employee benefits.' At the bottom, the page indicates it is built with Azure A I Search and Azure Open A I.
Pablo Castro retweeted
VoiceRAG: An App Pattern for RAG + Voice Using #Azure AI Search and the GPT-4o Realtime API for Audio from @pmc buff.ly/3Y6NIu8
Pablo Castro retweeted
Demo of real-time voice input for a RAG application:
youtube.com/watch?v=vXJka8xZ…
Code @ github.com/Azure-Samples/ais…
We had a great conversation with @boazashkenazy in this Shift AI podcast! We covered knowledge and reasoning, retrieval augmented generation, vectors databases, and of course Azure AI Search.
Excited for you to watch and listen to the latest episode of the Shift AI Podcast with @Microsoft Distnguished Engineer Pablo Castro.
Check out the @geekwire article that landed this morning.
geekwire.com/2024/shift-ai-p…
#ShiftAIPodcast #SimplyAugmented #GeekWire #Microsoft
Pablo Castro retweeted
Build hybrid search and sub-question querying with Azure AI Search and LlamaIndex!
In this tutorial, Khye Wei of Microsoft shows you how you can use LlamaIndex with Azure's AI Search to create powerful RAG applications, including
➡️ Hybrid Search
➡️ Query rewriting
➡️ the LlamaIndex SubQuestionQueryEngine
techcommunity.microsoft.com/…
Pablo Castro retweeted
Azure AI Search, a modern retrieval system for AI applications, now has a drastically larger vector capacity—without a price increase. Customers can now run RAG at any scale and at a lower cost: msft.it/6012c0xmQ #AzureAI
Now Azure AI Search can handle up to 11x the vector volume as a vector database, offers 3x-6x the storage capacity, and up to twice the performance, with no price changes. Ready for RAG applications at all scales. #Azure #AI #RAG
azure.microsoft.com/en-us/bl…
Pablo Castro retweeted
Semantic re-ranking with Azure AI Search is coming to @langchain4j !
Just finished my PR at github.com/langchain4j/langc… with the great help of @pmc 's team!!
I even added cool integration tests featuring French philosophers: github.com/jdubois/langchain…
Pablo Castro retweeted
Azure AI Search support is merged into @langchain4j !!
github.com/langchain4j/langc…
This is a first implementation with vector search, we’ll add more advanced features (hybrid retrieval + semantic ranking) soon!
Pablo Castro retweeted
@Azure AI Search has finally landed into @langchain 🎉
It includes full support for Hybrid search and Semantic reranker, RAG is now easier than ever on Azure with LangChain.js! 🦜🔗
And we'll continue improving it, send us your feedback 🙂
github.com/langchain-ai/lang…
"better late than never" is boring, how would Shakespeare would say it, according to GPT-4:
"In tardiness, there’s worth; far worse to never come."
Douglas Adams version:
"In the grand, inexplicable clockwork of the universe, where time is an illusion and lunchtime doubly so, arriving late is just the universe's way of saying that, in the grand scheme of things, you're probably early for something else entirely."
Vector search in Azure AI Search is now GA and ready for production workloads. Semantic ranker, to boost RAG applications retrieval quality, is also GA.
btw - Azure Cognitive Search is now Azure AI Search.
techcommunity.microsoft.com/…
