Real-world Industry Use Case: Uber's QueryGPT
What is QueryGPT?
QueryGPT is Uber's cutting-edge natural language to SQL tool, designed to bridge the gap between SQL knowledge and business concepts. It's an AI-powered system that generates SQL queries from natural language questions, making data access more efficient and accessible across the company.
The Challenge: 1.2 Million Queries per Month
To understand why Uber developed QueryGPT, we need to grasp the scale of their data operations. Uber's data platform handles 1.2 million interactive queries every month. This volume puts immense pressure on their data team, requiring not just SQL expertise and deep knowledge of internal data models and business concepts, but also, faster interpretation and execution to bridge the two together.
Before QueryGPT, writing a single query could take around 10 minutes. With their new system, Uber has managed to reduce this time to just 3 minutes – a significant improvement in efficiency.
The Evolution of QueryGPT
QueryGPT's development didn’t happen in one day, from its initial design to its current state took roughly 20 iterations! The initial version utilized a vector database to store database schemas and sample queries, along with similarity search to find relevant information. However, this approach faced challenges with complex questions and large schemas.
QueryGPT's Current Architecture
The current version of QueryGPT is a sophisticated system involving multiple AI agents working in concert:
… (check in the video for details!)
Continuous Evaluation and Improvement
A crucial aspect of QueryGPT's success is Uber's commitment to continuous evaluation and improvement. They've created an evaluation set with golden SQL queries to compare against the system's output, using metrics such as accuracy, table overlap, and successful runs.
Under the Hood: A Peek at the Code
Let's take a closer look at some key components of the system in the video:
Curious to learn more?
Join Professor Mehdi and myself for a discussion about this topic below:
What you’ll learn🤓:
🔎 Explain what QueryGPT is and why Uber developed it
🚀 Highlight the challenges Uber faced with data queries
🛠 Walk through of the evolution of QueryGPT's architecture (20 iterations!)
🎯Walk away with key learnings and implications for other companies (build vs. buy)
👇
Upcoming Course: Hands-on RAG Systems for Production - Before we dive deeper on this topic…
We have some exciting news! Our RAG live course is coming up soon, and as a way of giving back to our amazing community, we're offering you 15% off.
Just use this link:
We'd love to see you there! 🎉
In the course, you'll have the chance to connect directly with Professor Mehdi (just like I do 😉 in the videos), and you can even ask him your questions 1:1. Bring your real work projects, and during our office hours, we'll help you tackle your day-to-day challenges.
This course is for:
01 👇 𝗔𝗜 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 & 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀: For AI engineers/developers looking to master production-ready RAG systems combining search with AI models.
02 👇 𝗗𝗮𝘁𝗮 𝗦𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁𝘀: Ideal for data scientists seeking to expand into AI by learning hands-on RAG techniques for real-world applications.
03 👇 𝗧𝗲𝗰𝗵 𝗟𝗲𝗮𝗱𝘀 & 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗠𝗮𝗻𝗮𝗴𝗲𝗿𝘀: Perfect for tech leads/product managers wanting to guide teams in building and deploying scalable RAG systems
🛠️✨ Happy practicing and happy building! 🚀🌟
Thanks for reading our newsletter. You can follow us here: Angelina Linkedin or Twitter and Mehdi Linkedin or Twitter.
📚 Also if you'd like to learn more about RAG systems, check out our book on the RAG system: You can download for free on the course site:
https://maven.com/angelina-yang/mastering-rag-systems-a-hands-on-guide-to-production-ready-ai
🦄 Any specific contents you wish to learn from us? Sign up here: https://noteforms.com/forms/twosetai-youtube-content-sqezrz
🧰 Our video editing tool is this one!: https://get.descript.com/nf5cum9nj1m8
📽️ Our RAG videos: https://www.youtube.com/@TwoSetAI
📬 Don't miss out on the latest updates - Subscribe to our newsletter:
