Onehouse’s Post

AWS is doubling down on the data lakehouse with S3 Tables and SageMaker Lakehouse + Studio. It's great to see them moving in this direction, and while these feature launches show lots of promise, there are some potential drawbacks to the approach. I break down what this means for data teams in the latest Onehouse blog.

Looks like a good progress in making RAG work on enterprise data warehouses to simplify queries. AWS debuts advanced RAG features for structured, unstructured data https://lnkd.in/gWZEPS_F

Exciting news from the Snowflake Summit in SF! Neo4j just announced a game-changing partnership with Snowflake, integrating its native graph data science solution within Snowflake AI Data Cloud. This collaboration allows users to access over 65 graph algorithms instantly, all within the familiar SQL programming environment of Snowflake. Say goodbye to data movement and hello to advanced graph capabilities seamlessly integrated into your workflow. Check out the details. #neo4j #snowflake

Instant Insights: Snowflake’s Scalability Meets Amazon Q in Quicksight for Effortless BI Discovery Thats the title of the new blog that I have co-authored with Jagdeep Singh from AWS How do you go from loading Enterprise data into Snowflake platform, further enrich using third party data from Snowflake Marketplace and visualise the data in Amazon Quicksight. Not just that, use Amazon Q in Quicksight to ask natural language questions and generate visualisations from your data. For those of you who want to try this your self, follow along this blog step by step and try all of this in matter of minutes #AWS #Snowflake #LLM #GenAI #BetterToGether #QuickSight #SnowMarketplace Looking forward to lot more such collaboration blogs with the Ambassadors and Champions Roberto Baldoni Jagdeep Singh Gilbert Quinn Vinod Jaganathan Hasan Mirza Dan Hunt Henri Beaino Matt Marzillo James Sun Rithesh Makkena Michael Rhodes Yang Yang Mathew Zele Sudeep Babu Leo Park Harley Young Ronald Chung Hazirah Hasnan Jonathan Asvestis Kuldeep Venati Angela Koh Rolly M. Satchit Joglekar Vinut Shetty Prashant Yadav David John Chakram Kamal Manchanda Soujanya Konka Colm Mulholland Syarif Hidayatullah Muhammad Ardiyan Masayuki Osagawa Younghoon Jung Bharath Suresh Yang Yang

https://lnkd.in/gH49RUjQ #Neo4j® announced at #Snowflake's annual user conference, Snowflake Data Cloud Summit 2024, a partnership with Snowflake to bring its fully integrated native #graphdatascience solution within Snowflake AI Data Cloud. The integration enables users to instantly execute more than 65 graph algorithms, eliminates the need to move data out of their Snowflake environment, and empowers them to leverage advanced graph capabilities using the SQL programming languages, environment, and tooling that they already know. #graphdatabases #snow #datascience #datascientist #ai #artificialintelligence #ml #machinelearning #machinelearningalgorithms

📊🤖 Processing data for machine learning models: Check out this super interesting blog post by our tech lead Renato Cargnelutti, where he talks about the architecture we ended up building for one of our clients, using #AWS SQS, #AWS Lambda, #Airflow, #DBT, and #Snowflake to process ML model data. 🔗 Read the full post here: https://lnkd.in/dB5JnhPh

Check out our latest blog and learn about unleashing the power of data governance and no-code machine learning with #SageMaker #Canvas and #DataZone. In this blogpost, we show how the DataZone integration with SageMaker Canvas empowers #ML teams to perform no-code ML with the governance capabilities. Unlock the full potential of #ML and #genAI while maintaining control and oversight over your enterprise data assets. https://lnkd.in/g5kiVzXY Siamak Nariman Ajjay Govindaram Huong T. Nguyen Lauren Mullennex Yueying Cui Derek Young Nausheen Sayed Divakaran (Diva) U. #GenAI #AWS #SageMaker #Canvas #DataZone #ML #AWSBlog

Snowflake became the latest company to build their own LLM, here is whats different about it. Snowflake Arctic is focused heavily on the types of enterprise use cases that Snowflake already serves. They trained it in three phases, gradually increasing the ratio of Code, SQL, and STEM content until it made up 35% of the total training tokens in the last phase. Architecturally, they took Mixture of Experts (MoE) to the extreme. In this format, the model is divided into a number of 'experts', with only a few being active at any one time. Previous implementations had as many as 8 experts, Snowflake kicked this up 16x to 128 experts. This means they built a 480B parameter model, but at inference only 17B are active at a time. Having such a wide MoE means that you can get a great performance/compute ratio, but only at scale. The recommended base setup is 8x H100s, although they claim a heavily quantized version can fit on a single GPU. Makes sense for Snowflake's serverless business model, but tricky for anyone looking to self host. https://lnkd.in/en-8SGwh