With the advancements in in GenAI, as someone who works on a social media product, LLMs felt like they were calling to me.
We had already set out to build some feature involving LLMs, like trendline analysis, spike analysis, and topic clustering.
But i felt there was a calling towards something that was more bespoke, the user is combing through a huge amount of data with a question in mind, or even trying to find a question to ask, how can i help him with that.
The requirements were that it shouldn't be just a generic chatbot interface that isn't capable of showing the user the full context of the analysis, it needed to be very specific to the type of analysis we have.
We also though of building a full on Reporting Agent, capable of building reports on it's own, but the question remained: what are the basic components that that this agent will use to generate these insights, what are the tools that'll make him able to ask questions and craft them into a coherent story to tell.
I went ahead and built the Insights Workbench, my first iteration on an answer to this question, here are the basic functionalities it can do for you:
- Get you all data that is relevant to you question (Yes, relevant not just similar).
- Discover how data breaks down in relation to some criteria (e.g. attitudes in AI).
- Classify your data into distinct classes.
- Make an analysis based on a free text question.
While initially developed as an internal tool to iterate on, we were able to imagine it as real integrated feature inside the app, which is now under the works.
The code is publicly available for now and you can check it out here, you can test on your own data provided it matches the schema.
Best of luck!
Top comments (0)