Yogesh Kandlur

Whoever claims building RAG systems is easy has never built one at scale. 🤯 There is a significant gap in the literature and existing frameworks. RAG architectures are designed for POCs, not for hundreds of thousands of documents or supporting tens of thousands of users. Designing these systems requires a cross-functional team with software engineering and data science skills blended with a product mindset. Earlier this summer, I spoke at the Databricks AI Summit, highlighting guiding principles we are taking at CVS Health for building enterprise-grade RAG systems. Every company is unique, and while there is no one-size-fits-all solution, here are three principles that will help you on your journey: 👉 Spend time on discovery—It's tempting to dive in and start building immediately, but you may get lost in the desert without doing your due diligence and understanding what delivers the most ROI. Our team spent approximately a quarter on discovery. 👉 Design for modularity—The LLM space is changing rapidly, and you want to build reusable components that can easily be swapped out or upgraded when something better comes along. 👉 Humans + Machines—Technology alone can not solve the knowledge search problem. Develop processes and governance to help teams curate the data, which will be ingested into your system. These lessons are just the tip of the iceberg. Watch my full Databricks presentation and leave your comments and questions below to learn more tips and tricks. #RAGSystems #EnterpriseAI #DataEngineering #MachineLearning #AIArchitecture #AI #DataScience #LLM #SoftwareEngineering #AIAtScale #KnowledgeSearch #TechLeadership #AIFrameworks #AIInnovation #DatabricksAISummit #ProductMindset #CVSHealth

62

5 Comments

🚀 Ridge Regression: Your Guardrail Against Overfitting 🛡️ Imagine you're a retailer predicting sales for thousands of products. 📊 Without Ridge Regression, your model might overfit to noise in the data, like one-off promotions or outliers in specific SKUs. This could lead to inaccurate forecasts and overstocking! By applying Ridge Regression, you can control overfitting, ensuring your model focuses on broader trends, like seasonality or customer preferences, rather than just noise. The result? Smarter inventory management and higher profitability! Curious to learn more? 👉 Read here: https://lnkd.in/g8RbkyZJ #RetailTech #DataScience #MachineLearning #RidgeRegression #AI

7

Common Agentic workflow patterns #llm #llms #agenticworkflow

1

Improving retrieval is the cornerstone of any successful Retrieval-Augmented Generation (RAG) system. While numerous RAG methodologies showcase impressive potential, a foundational technique—Hybrid Search, which combines Dense Vector Search with Keyword Search—consistently delivers superior performance across various queries. For those venturing into RAG for the first time, starting with Hybrid Search as a benchmark is highly recommended. Its ability to leverage the strengths of both semantic understanding and exact keyword matching makes it a robust choice, laying the groundwork for deeper exploration into more advanced RAG methods. #llm #llms #RAG

12

With self-serve analytics, you can empower everyone in your company to go from data to decisions. Our VP of Analytics Adam Kinney explains how we do it at Mixpanel 👇 https://lnkd.in/dUeb4mDe

6

I am always fascinated by the aggregated benchmark data we receive from the Brightflag platform - being a data geek each metric always make me want to dig deeper and ask more. Here is one metric that will help you evaluate associate rates quickly. Associate rates peg very predictably to partner rates. You expect they should be around 65% of the partner rate for any given firm and practice group. So here is a quick formula for evaluating associating associate rates: 1. For each associate calculate their rate as a proportion of the the average partner rate at the practice group or of their managing partners rate 2. If they are more than 70% of the partner rate then they are an outlier 3. This allows you to ask questions of the proposed or billed rate: - Are they senior level associates? - How many years of experience do they have? - Are they the most appropriate associate for the work billed? Are there any other simple but useful metrics you use for evaluating billed or submitted rates?

25

10 Comments

Kumo helps data scientists build highly accurate predictive AI models that predict user and customer behavior by combining graph learning over enterprise data and generative AI models trained on public data—all entirely within Snowflake. If you want to create better machine learning models faster with Kumo, try our Snowflake Native App now. 🔗 You can request a free trial at the following link: https://hubs.ly/Q02HgQ1P0. If you need additional support and more details, contact us. We'd love to get in touch with you and help you query the future: https://hubs.ly/Q02HgGvG0. #AI #MachineLearning #PredictiveModels #DataScience #GraphLearning #GenerativeAI #Snowflake #TechInnovation #KumoAI #FutureOfData

9

Unlock the power of Snowflake for your multi-cloud strategy! Whether it's cross-cloud compatibility, secure data sharing, or optimized costs, Snowflake’s unified platform simplifies data management, scales effortlessly, and ensures high availability. Learn more about how Snowflake can elevate your multi-cloud approach and drive business innovation in our latest FAQ:  https://lnkd.in/gyPdk67P  #MultiCloud #DataStrategy #Snowflake

2

Ramp's data team is growing! We just opened up an analytics engineering role with a focus on solving quantitative problems across risk and capital markets. https://lnkd.in/es73XKk8 There are a few reasons why this role is exciting: The mission, opportunity and product. Corporate spend management is frustrating, reactive, and time consuming. Ramp gives finance teams visibility, control, and the ability to automate. Ramp enables companies to save money, in an industry where cards and spend management software have not historically been linked. You can see anecdotes all over from satisfied customers and companies. The analytics challenges. There’s a wealth of first party transaction data, second party spend and accounting data, and third party data from across the financial ecosystem. The opportunities are fascinating, and it won’t work without a world-class analytics platform. Finally, the team. James Rizkallah, Ryan Stevens, Karen A., Kevin Chao, Ryan Foss, and William Petrie have built an incredible foundation at the intersection of capital markets and analytics. The speed of execution speaks for itself. There's a lot of very cool problems and a ton to build! Come join us 🚀🚀

93

3 Comments

In my view, a data product is a concept that involves structuring and managing your company’s data in a manner that promotes usage and actions. It’s akin to adopting a product-oriented approach to your datasets, viewing them as valuable assets: 1. It explicitly identifies the personas it caters to, with a strong emphasis on the end users! 2. It tackles business use cases pertinent to the selected domain. 3. It features a highly user-friendly application interface to cater to the typical needs of the personas. 4. It distinctly prompts actions rather than merely offering insights. 5. Finally, the data product should encompass a unified data layer that includes modelled data and accompanying metadata. It was great conversing with Brian T. O'Neill! Do listen to the podcast. #TGIF

49

4 Comments

ChaosSearch, as a Databricks Technology Partner, brings ELK use cases to Databricks ecosystem enabling centralized log and event analytics for observability and security. 🔒 Deliver on the best of Databricks (open Spark-based data lakehouse) and ELK (efficient search, flexible live ingestion, API/UI) via ChaosSearch on Databricks. Data ingested into Delta Lake format, all compute in Spark via Databricks and with the best of ChaosSearch’s hunting and Databricks’ AI/ML. Explore the power of ChaosSearch in your Databricks environment: https://lnkd.in/dSAXRgEF

4

Mastering the art of metric selection will have compounding returns in your experimentation journey. I am hosting a webinar next week to share my learnings on metrics. Join if you’ve ever faced these challenges in A/B testing: - Your experiments lack statistical power for critical business decisions. - Your team hesitates to experiment due to fears of hurting key metrics. - You struggle with fluctuating, unreliable results. - You want deeper insights but don’t know how to extract them. Register here: https://lnkd.in/gszag6HK #experimentation #metrics #webinar

51

We build multiple custom analytics solution and help companies scale with data One of the solution brought way over 10m in pure revenue The others decreased churn and uncovered truth bottleenceks they weren't able to see before All this happened because they done these 3 steps 1) Deep data strategy Data strategy hence the name is assessing how to use data to align with your business goals Realizing what is priority what is not comparing to business value vs feasibility This can determine if the solution gone generated you ROI or be useless Only possible to do this is by deep expertise in business and data 2) Unification Breaking data silos and having data organizad, accurate in one central place called data warehouse.  This is crucial if want to generate insights and be data driven you need to see and get cohesive view. For example if want to determine a problem in churn you won’t just look at CS metrics but you will look at sales, marketing and finance. 3)Adoption & Iteration Business changes fast, thats why your data has to adapt to your needs. Not possible to scale to 10m with data maturity being at level 1. This is not a one time thing and its done. If you think it is you can try it and will probably stop using the solution or stop being data-driven in 2 weeks which will lead to not good things. Iterative approach sprint > 4) Actionable Insights All of these results were able to achieve because they had metrics, custom metrics presented and created in way that forced them to be better everywhere. To create actionable insights you need •Story - Why is this important?  •Context - What is main topic to focus on?  •KPI - Provides additional context Only able to achieve that by deeply understanding the business, user, industry and of course the data (technical part) If looking to fill out that gap covering both of those worlds-> Check aztela(dot)co

4

2 Comments

https://lnkd.in/gYbfckU7 #RecommendationEngine : I have analyzed Big Basket's customer buying behavior data and created a recommendation engine that suggests items you might be interested in. this is programmed in R language and hosted in Shinny. Check it out! I would greatly appreciate any feedback.

11

3 Comments

The Charlie Health #datateam makes accurate data insights available instantly. They do it with TigerGraph, Snowflake and Hex. See how on Dec 12: https://lnkd.in/edpueg-y

15

ZoomInfo data is now available in Databricks' AI/BI Genie — unlocking a whole new way to interact with your data. 💬 No more sifting through endless dashboards, waiting on data experts, or wrestling with SQL. With AI/BI Genie, getting insights is as easy as asking a question. This conversational AI empowers business teams to engage with their data using natural language. By combining ZoomInfo data with Genie's generative AI, non-technical users can ask questions about their enterprise data just like they would an experienced coworker — and receive accurate, real-time answers. Learn more about Databricks and ZoomInfo here: bit.ly/4f53MBT

129

6 Comments

Let me know if you need help registering - #CallTheBevForAnalytics Solutions & Advice Join #TheBeverlyOrbit

8

KL Databricks How to Series: Automate Notebooks Using Databricks Jobs! In this quick demo, we’ll guide you through automating notebooks using Databricks Jobs! From navigating the Job Runs section to configuring tasks and clusters, we’ll show you everything you need to know. You’ll learn how to set task dependencies, schedule jobs, and monitor task execution, making it easier than ever to automate your data pipelines. Watch now for hands-on insights! For any inquiries, contact us at [email protected] #ETLAutomation #DatabricksJobs #DataEngineering #BigData #CloudComputing #DataOps #TechInnovation #DigitalTransformation #DataSimplified #TechTutorial Kadel Labs

11