Loading Events
  • This event has passed.
6 Jun 2024 18:30 - 21:30
datacraft –
3 Rue Rossini
75009 Paris, France
+ Google Map


Start with drinks and food : 6:30pm

Speaker: Fabio Buso, VP of Engineering at Hopsworks

Title: Solving Stockholm commuters pain using LLMs with Hopsworks
Abstract: Retrieval augmented generation (RAG) can be used to personalize LLMs interactions by injecting a prompt to the user query. Vector indexes have been the most common way people build RAG pipelines by indexing and retrieving unstructured data such as text documents. Vector indexes, however, struggle with real-time data and are not ideal to store and retrieve structured data. Feature stores can be used as a RAG pipeline data source for real-time structured data source. In this talk we are going to explore how to use the Hopsworks feature store to combine document-based RAG pipelines with real-time structured data from the feature store. We’ll do so by building a LLM based application to plan my commute on the Stockholm commuter rail.
Speaker: Thaïs Denoyelle – Data Scientist at Datacraft
Title: Exploring Polars: A Critical Analysis of the New High-Speed Python Library for Data Analysis
Abstract: This session focuses on Polars, a Python library gaining attention for its high-speed data analysis capabilities. Through a critical analysis, we’ll delve into Polars’ architecture, functionalities, and performance, comparing it with established tools like Pandas and Apache Spark. Attendees will gain insights into Polars’ features, potential applications, and implications for data-driven industries. Whether you’re a data scientist or enthusiast, this session offers an in-depth examination of Polars and its impact on the landscape of data analysis.
Speaker: Jean-François Macresy – CTO/CPO at Videho
Title : ML Ops for video content
Abstract: Video content has multiple particularities. It is heavy, difficult to parse, and multidimensional as it is composed of frames stacked together. This talk will dive into the specificities of manipulating such content in the context of a video ML platform, and how to architect it to be scalable, monitored and easily modifiable.
< All past workshops