Dataiku for AI & Machine Learning

Summary

Highlights

• Dataiku boosts data science efficiency and innovation ⚡
• Predictive maintenance enhances machine performance 🛠️
• Seamless data access and preparation from various sources 🚪
• Advanced EDA and modeling using Jupyter notebooks 📒
• NLP tasks made easy with integrated LLMs 🧠
• Efficient model comparison and benchmarking 🔍
• Robust model deployment and monitoring 🚦
• Centralized governance and automation improve collaboration 🤝
• Fast and effective project execution 🚀

Key Takeaways

• Dataiku accelerates data science workflows and boosts innovation ⚡
• Access and prepare data easily from diverse sources 🌐
• Leverage robust modeling and app development features 📊
• Seamless integrations with favorite third-party systems 🤝
• Ensure robust model performance and stakeholder communication 📈
• Centralized governance for data and analytics projects 🏢
• Efficient deployment and monitoring of AI services 🚀
• Automation and orchestration streamline model management ⚙️

Overview

Chapters

• 00:00 - 01:30: Introduction to Dataiku The chapter introduces Dataiku, a leading AI and machine learning platform, emphasizing its role in helping data science teams focus on essential tasks. It highlights how Dataiku can assist teams in meeting internal customer demands by streamlining their workflow and reducing the time spent on non-essential tasks.
• 01:30 - 03:00: Exploring Data with Dataiku The chapter titled 'Exploring Data with Dataiku' discusses how Dataiku serves as an essential tool for data science teams by boosting workflow efficiency and reigniting innovation enthusiasm. It covers the complete data science lifecycle, from data access and preparation to modeling, app development, and operationalization. Dataiku is portrayed as versatile, supporting traditional machine learning, generative AI, and hybrid approaches. A practical example is provided where a team aims to improve machine efficiency and predict potential failures through predictive maintenance using Dataiku.
• 03:00 - 05:00: Data Analysis & Modeling The chapter 'Data Analysis & Modeling' focuses on the proactive maintenance of company machines to minimize downtime and improve performance. A data scientist, part of a team of 10, discusses the creation of a model for real-time predictions for machine maintenance. The first step involves accessing trusted data from various sources such as databases and cloud storage.
• 05:00 - 08:00: Model Deployment & Monitoring The chapter discusses the process of exploring data, particularly sensor metric data, as part of model deployment and monitoring. Initially, exploratory data analysis (EDA) is conducted using data tools like data IU's analyze window and statistics cards, providing a quick overview. Further, deeper investigation is carried out using Jupyter notebooks for programmatic data exploration, allowing for comprehensive understanding and further analysis.
• 08:00 - 09:30: Orchestration & Automation The chapter on 'Orchestration & Automation' discusses the integration of necessary dependencies and packages within a notebook on elastic compute to ensure scalability. It highlights the ability to quickly identify trends and the decision to incorporate historical maintenance records from field teams for thorough analysis. Utilizing natural language processing (NLP), the chapter demonstrates how unstructured data can be efficiently managed. Moreover, it emphasizes the use of flexible connections to both hosted and local large language models (LLMs) via the data IOU LLM mesh, ensuring every call is logged and checked for toxicity and personally identifiable information (PII).

Dataiku for AI & Machine Learning Transcription

• 00:00 - 00:30 [Music] in data science time is of the essence but many teams find themselves bogged down in non-essential task and therefore struggle to meet internal customer demands enter data ioup the leading Ai and machine learning platform that's
• 00:30 - 01:00 accelerating workflows and helping data scientists ReDiscover the joy of innovation from data access and preparation to modeling app development and operationalization data ioup is the go-to solution for data science teams whether dealing with traditional ml generative AI or a combination of both let's see an example of dat IU at work my team is looking to enhance machine efficiency and forecast potential failures through predictive maintenance
• 01:00 - 01:30 this approach enables proactive maintenance of our company's machines minimizing downtime and improving overall performance after a meeting with business stakeholders I one data scientist on a team of 10 AM tasked with creating a model for real-time predictions for proactive machine maintenance using data IU's data collection feature the first thing I do is get my hands on the data I am able to access trusted data from various sources including databases and cloud storage I
• 01:30 - 02:00 can even look at the scalable Upstream pipelines that generated the data set being unfamiliar with the data the first thing I'm going to do is explore it to get a quick view of the sensor metric data I'll use data IU's analyze window and statistics cards to do quick Eda but that's not enough I need to explore further no problem I'll open up a Jupiter notebook and continue to investigate programmatically with all
• 02:00 - 02:30 the necessary dependencies and packages readily available in the notebook running on elastic compute for scalability I can quickly identify Trends I've decided to bring in historical maintenance reports from field teams to help with the analysis I'll use natural language processing on these unstructured records with data I coup these sorts of NLP task are a breeze with flexible connections to hosted and local llms right through the data IOU llm mesh with every call logged and checked for toxicity and pii without
• 02:30 - 03:00 any additional setup I can incorporate this data super quickly through code or visually and dat IU's prompt Studios with a solid understanding of the data I am now able to confidently hand over the project to my colleague the lead data scientist on this project to experiment with different models using dat IU's native support of ml flow open source Frameworks all metrics parameters and artifacts are seamlessly saved in the project she compared different model
• 03:00 - 03:30 versions and even benchmarked against the dat i coup automl model that another colleague who has since left the team had created previously once satisfied with the model the lead data scientist save the code and model to dat IU's flow as reusable recipes ready for orchestration it's worth mentioning that for this project the lead data scientist leveraged trusted code and functions from her favorite get Repository
• 03:30 - 04:00 seamlessly connected to data IOU however some other members of our team prefer to use an IDE like vs code with dat IU our different preferences are not a problem our favorite idees are also accessible directly within dat IU's code Studios and with dat iou's AI code assistance we ensure our code is robust we can each use the tools we like best but all in one centralized place so that we can work faster together although for this use case I trained the model directly in
• 04:00 - 04:30 data IU surfacing endpoints from external models or importing models trained elsewhere is just as quick and streamlined thanks to data IU's Integrations with all of our favorite thirdparty systems like data bricks AWS Azure and gcp no matter where we train or deploy models dat IU is our Central hub for Ops monitoring and governance whether a day I coup model ml flow model or a third party model saved models come with default explan explainability and
• 04:30 - 05:00 performance information to help with validation and stakeholder communication that's one of the many reasons dat IU is our Central hub for example here I've saved some of my favorite results to a dashboard for easy sharing with others on my team as well as a few of our ml Engineers with stakeholder buyin on our model how are we going to monitor it over time in dat i coup monitoring is extremely simple and done from the start without the need for deploying first with data who's model evaluation store
• 05:00 - 05:30 we can set up access monitoring automatically logging metrics related to data performance and prediction drift these metrics serve as automation thresholds triggering alerts and retains through data ioup automation scenarios and if there is an unforeseen issue dat coup saved models makes retraining and virgin roll back a breeze we're confident that our model is high performing and built in a responsible way with everything we need to alert our team about drift over time now it's time
• 05:30 - 06:00 to deploy a live API endpoint so it can be used by our Factory management software for proper team intervention did IU's deployer is the One-Stop shop for deploying projects and API Services managing their life cycle across environments and monitoring their health now my team hands off to our ml engineer who will deploy our new API service into production dat i coup automatically enforces our organization Technical and governance checks in this
• 06:00 - 06:30 case we can test check our [Music] scenarios model deployment is just a few clicks away in data ioup but it also ensures the right approvals and sign offs before critical use cases enter production dat IU govern is a centralized Watchtower over our organization's analytical projects and models ensuring we're not putting the company at risk with any of our projects or deployments for our data sign team
• 06:30 - 07:00 upholding a strategic overview of the entire data and analytics landscape at our company is imperative dat IU's model and project registry provides a centralized way to see all the AI projects with performance metrics and summaries for leaders and project managers once we have sign off and our model is deployed we can easily monitor the endpoint activity and health be it on the model drifting or a technical deployment issue
• 07:00 - 07:30 once an issue is detected we have a clear breadcrumb Trail back to the related drift check model version and project to further streamline the process our data science team uses data iou's orchestration and automation capabilities we set up scenarios for data pipeline refreshes and automated retraining based on drift ensuring the model remains accurate and reliable over time with alerts and visual job logs we feel confident our API is executing
• 07:30 - 08:00 correctly thanks to dat i coup our team successfully tackled this complex predictive maintenance use case creating a high- performing responsible model that enhances machine efficiency and predicts potential failures the streamlined process and efficient collaboration not to mention quick time to Value made the project a resounding success thanks for watching