Gemini for Data Scientists and Analysts

Key Takeaways

[Music] [Applause] this is a data professional they might be a data analyst data engineer or data scientist we'll refer to them as the data Pro let's learn how you can use big query Gemini and vertex AI to analyze your data to categorize new customers or gain insights like predicted product sales later you'll use Gemini vertex Ai and big query to generate useful next steps for a marketing campaign they've recently joined a new company and part of their role requires them to turn their team's manufacturing plant dashboard from a reactive one into a proactive one so that their team can anticipate issues with equipment the data Pro is here to help ensure that the equipment is functioning properly they do that by using data to create reporting mechanisms that predict when maintenance should occur and Gemini is here to help the data Pro the first step is to open the manufacturing plant dashboard powered by looker all of the plants real time stru Ed and unstructured data is loaded into Google Cloud analyzed and presented here they can identify that one of the plants is having some issues so they click into it to explore immediately they notice that throughput has been below Target for a number of hours starting at 2: p.m. the data Pro decides to use Gemini in looker to learn why throughput is down in the chat window they prompt Gemini why did the throughput drop at 2 p.m. Gemini responds with the production line that throughput dropped four and because it knows the context of the situation Gemini has gone one step further and suggested additional questions to ask like this one which metrics correlate with the dip from the response the data Pro realizes that there are some equipment issues Gemini has gathered from routine Diagnostics some machine statistics that our Pro will use to start communicating the physical fix to technicians this is really helpful for our data Pro now our data Pro can understand why they were tasked with making this dashboard proactive instead of reactive they need to build a preventative maintenance model to predict future downtime and avoid it to build and ship a data model they start in bigquery studio a simple unified environment where they can build anything they want including in Python first they determine what data already exists so they don't have to start from the beginning there are two data sets one in Google cloud and another in a different Cloud they notice a throughput table that might be a promising starting point they open it to validate the usefulness and quality of the data when they open the table they use the data profile tool to perform some light exploratory data analysis they also note that there are automated quality checks in place which is a good sign so they know this data is actively maintained there is even an insights tab where Gemini has already analyzed the data and how it's being used and proactively suggests insights for business questions that our data Pro might want to answer they decide to open and run one of the queries in big query to further explore the insights they appreciate how helpful these insights are to to build their understanding of this data quickly it seems that another data set also has some reference data so they'll pull that into the model too because they're building multiple ml projects they decide to do this in a python notebook so collaboration is easier Gemini supports SQL generation and completion so they ask it to use the multicloud data to help build a model that can predict throughput based on the attributes currently available in the python notebook itself they prompt Gemini by clicking Plus Code and typing in a prompt to create a Logistics regression model and just like that the SQL is ready that saved our data Pro a lot of time it also interconnected the data across multiple clouds by using a simple SQL join in real time without the need for complex operations or processes here Gemini helps our data Pro code faster discover the data answer business questions and generate insights amazing our data Pro wants to ensure that the model will actually run on real-time data because it was trained on a table not a data stream first they use a few lines to verify that the model is actively running against the production data and with predictions so the model is ready predictions are a good start but the data Pro wants to go further and actually help the plant operators by giving them recommendations for how they can keep the plant working optimally luckily there is some more data available this time unstructured inspector reports that may help they use retrieval augmented generation which means that they will use their own data and an llm a large language model to build an agent using tools to connect to their data these tools will help improve the reliability of responses from the llm and reduce hallucinations which happen when the llm produces inaccurate responses sounds like a lot of work but they're going to do it in a few lines of code first they'll bring AI to the data in big query they use llms powered by vertex Ai and bigquery's new ml inference engine to connect to all these amazing models summarize the reports and create vectors or numeric representations of their text to help with searches and groupings next they pass a question or prompt into the llm to get the vectors and summary then they use the summaries to ask the llm for recommendations to get the predictive actions and the results are displayed this would have taken our data Pro a lot of time and effort to do if they use traditional data science methods here they did it with a few functions and in a couple of minutes by using the power of generative Ai and all in big query where their data remains secure now the pipeline is deployed here are the pipeline results back in the plant dashboard where the predictive results are now displayed our data Pro can even set actions by using looker so that the right people are notified to take action on the new recommendations our data pro has succeeded in helping the equipment monitoring dashboard move from reactive to proactive and with all that time saved they can now finally get around to playing the ukulele thanks Gemini what will you build [Applause] [Music]