By now, it’s clear to everyone how AI has increasingly begun to reshape our daily work lives. Although we are still in the early stages of this journey, it is clear that the future leaders will be those who effectively harness the power of AI. Therefore, it has become more crucial than ever to embark on this journey, find the right methods to leverage AI, and begin integrating it into the core of our business operations. If the terms related to GenAI are still new to you and you need a refresher, make sure to review this user-friendly summary with examples here.
As with any endeavor, architecture plays a crucial role in the realm of AI as well. You can read more about this in detail in a previous Ikidata blog post here: GenAI Architecture. To summarize briefly, seamless integration and data accessibility are imperative when leveraging AI. If solutions exist in isolation and data is not readily available, a functional data architecture remains a mere dream. This is why technological choices and platforms are of paramount importance. And as it happens, there is already one standout option on the market - Databricks.
Databricks - Data Intelligence Platform
If you are not familiar with Databricks, they pioneered the data lakehouse, combining data warehouse and data lake capabilities to manage both structured and unstructured data for analytics and AI. In November 2023, they launched the Databricks Data Intelligence Platform, integrating MosaicML’s Generative AI to enhance data utilization.
With Databricks, you can easily adopt the latest GenAI tools and solutions, as it provides a comprehensive ecosystem for both data platform development and advanced analytics. In this blog post, we will explore the out-of-the-box GenAI solutions offered by Databricks, which can elevate your work efficiency to an entirely new level.
Databricks Assistant
Everyone who has seen Iron Man has dreamed of having their own AI assistant. A selfless helper like J.A.R.V.I.S., who knows everything and processes requests instantly. While we haven't quite reached that level today, GenAI Assistants are becoming relatively proficient in basic coding and data tasks. Databricks also offers its own assistant, which performs well for fundamental needs. The most important difference from other similar tools is that it is seamlessly integrated and embedded into the Databricks UI. This makes it a highly efficient option.
The Databricks Assistant is automatically enabled and follows you wherever you go — both for better and for worse. In the illustration, you can see the icon that allows you to activate it. It works excellently as a coding assistant for basic tasks in Python, Scala, and SQL. However, when it comes to more complex implementations, it struggles to keep up. Despite this, it allows individuals without a coding background to create functional coding solutions without writing a single line of code themselves — it's incredible! This way, business professionals can get involved and dive into the fascinating world of data platforms.
Noteworthy features include predictive text input and quick fixes. While these can be time-savers at their best, they often feel overly aggressive and occasionally get in the way due to overzealous hallucinations. It's essential to remember that these features aren’t one-size-fits-all; finding your own way to integrate AI assistants into your workflow is key to maximizing their benefits. Assistants should enhance and support your work, not dictate how you operate.
AI/BI Genie
AI/BI Genie is a conversational tool that empowers business teams to interact with their enterprise data using natural language. Built on generative AI customized to your organization's data and terminology, Genie continuously adapts through user feedback. It enables users to ask questions as if consulting a knowledgeable colleague, delivering accurate, context-aware answers instantly. With seamless integration into Databricks SQL, Genie provides fast, reliable insights beyond traditional dashboards, all while maintaining governance and security through Unity Catalog.
Genie offers an intuitive Text2SQL solution, translating your natural language prompts into SQL queries, fetching data from selected datasets (you control which datasets Genie can access), and summarizing results in natural language. One standout feature is Databricks' transparency: you can see exactly which SQL queries Genie used. You can also refine Genie's performance with preliminary prompt engineering by providing text instructions and example SQL queries, and improve it further by giving thumbs up or down based on answer quality.
After some fine-tuning, Genie delivers quite good results. But the real game-changer is its seamless integration with AI/BI dashboards. Imagine having your own chatbot or assistant seamlessly embedded into your Power BI dashboards — something you’ve probably dreamed of. Now you can achieve it with Databricks AI/BI Dashboards. With just a single click, you can elevate your data analysis experience to a whole new level, leaving traditional dashboard methods far behind.
For more details, check out the full article on getting started with Databricks AI/BI Genie!
Model Serving Endpoint
Once you've spent some time working with the tools Databricks offers, you'll likely want to start building your own GenAI solutions. The answer to this lies in model serving endpoints, which allow you to host your own LLM models via Databricks or integrate with external models. In this setup, Databricks acts as an intermediary, providing powerful management capabilities through Mosaic AI Gateway. You can read more about Mosaic AI Gateway here.
In the Serving tab, you'll find options for setting up serving endpoints. Databricks provides Meta's llama-3-70b-instruct model as a default, with usage-based pricing, making it an excellent starting point. You can also create serving endpoints for external service providers like Anthropic or (Azure) OpenAI. This is a quick and efficient operation that can be fully automated. Think of an external model as a Python wrapper that acts as a bridge to call the underlying LLM model.
But why use Databricks' model serving endpoint instead of a direct connection? The answer lies in Mosaic AI Gateway. It allows you to effortlessly set rate limits for each endpoint, even at the individual user level. Additionally, guardrails are easy to implement, ensuring safe usage. The biggest advantage, however, is the automatic inference tables, which provide robust monitoring of model usage. You can learn more about leveraging inference tables here.
Playground
Now that you have direct access to an LLM model in Databricks, you'll want to make the most of it. With the new Playground feature, you don’t even need to write any code. You can easily run models directly through the interface, and best of all, you can run them side by side for A/B testing, allowing you to see real-time differences in their behavior. This feature works seamlessly, enabling you to experiment with various system and user prompts to optimize your outcomes.
In addition, you can kickstart your GenAI Agent journey by granting LLM models access to functions (limited to those registered in Unity Catalog). This dramatically enhances functionality and efficiency. On top of that, you can export configurations directly to your own notebook, enabling further development and the creation of dedicated GenAI Agents tailored to your specific needs. It’s an excellent way to take your GenAI projects to the next level!
Start Your GenAI Journey with Databricks
This is merely a glimpse into the possibilities of GenAI within the Databricks ecosystem, but it should give you an idea of the immense potential. The whole data ecosystem is set to change permanently, as we can now truly deliver citizen GenAI services through Databricks. This means Data Scientists are no longer bottlenecks, allowing business teams to transform their brilliant ideas into tangible data-driven solutions — without getting stuck in endless meeting marathons. And this is just the beginning. Now is the best time to start your GenAI journey with Databricks!
The future looks incredibly promising!
-𝐾𝑟𝑎𝑡𝑡𝑖
Related Posts
