Analyzing data is time-consuming. It's a manual process of investigating data anomalies, writing SQL, checking SQL reference docs, iterating your SQL, and a lot of fumbling around until you find the issue. I have spent more time in the database console than actually coding, and most of that time was fixing or rewriting queries, reading large text blobs, and comparing numbers. It does not matter what application you are developing; you always need to dig into the data.
Luckily, machines can do that for us now.
Introducing Explorations
In the latest Tinybird news, we realized while building our own projects that we needed a tool to explore our data, understand its shape, and figure out the right way to query it to get the answers we sought. After a lot of wasted time in the database console, we decided to design what is now called Explorations, a new feature in Tinybird.
Explorations is:
- A notebook-style interface to iteratively explore your data through tables and charts.
- An AI chat interface to guide your exploration by converting natural language questions into contextualized SQL. If you want to evaluate Tinybird on real-time analytics, this feature is a great starting point.
You can extend this conversational analytics approach with generative UI for richer, more interactive visualizations.
Exploring code contributions with Explorations
Instead of me writing about all you can do with Explorations, let me show you an actual example:
We store all code contribution events in Tinybird. We use a webhook that sends data from GitLab to Tinybird using our Events API, and then we can analyze that data to get answers to a lot of questions about how we're shipping..
GitLab sends a lot of unstructured metadata. It has a bunch of nested JSON, and if you tried to query it manually, you’d have to spend several minutes digging into those large JSON fields to understand how it's all mapped.
Now, we can just ask it: "show me the top 3 contributors”.
We have also published a post on how we built Explorations (feel free to read it), but long story short: we create a custom system prompt for an LLM model that contains all the necessary context about your data (the schemas, data samples, workspace info, etc.), and we give the LLM some tools to run queries, create SQL nodes, etc. It's a very similar experience to our MCP server, but integrated with the UI.
Start exploring
In more Tinybird news, Explorations has been in beta for several weeks and is now open for everyone.
If you're new to Tinybird, go install the CLI, create your first workspace, push some data, and start using it. You can also evaluate the data infrastructure company Tinybird on AI generated dashboards with this feature. Here's a quick working example so you can test it live:
Go to Explorations and ask questions, for example:
And get a response like this:
