Inspiration

Last week I had a call with the team in my project about new requirements from the business regarding displaying different business data. Again. New version. It is always a compromise between different departments about dashboards, page layouts etc. I said, "There should be AI Agent that prepares dynamic dashboards and then each User will see desired stuff.". One girl replied: "Please do not say it to the business.". I will not say it to them. I will show it to them.

What it does

dashpers allows to create personalized business dashboards through natural language commands.

How we built it

Stack: lovable (React, TypeScript, Tailwind), fal.ai, pica, render, IIElevenLabs, supabase, edge functions. Used Anthropic framework of effective ai agent - Building effective agents

Challenges we ran into

  • I had to run external Node.js server to be able to use Pica SDK, it was not possible to use it via supabase edge functions,
  • max stack limit calling IIElevenLabs,
  • lovable changed unnecessary code from time to time.

Accomplishments that we're proud of

In quite limited time and resources (solo) I was able to provide some features that I am satisfied with. I was struggled at the beginning with some tools that I did not use before (probably I could deliver more with Cursor), but finally I am happy about the results.

What we learned

This time instead of Cursor I decided to use new tech stack: lovable, pica, fal.ai - I did not use it before. It was fun!

What's next for dashpers

Publish quick demo for feedback and talk to users. Regarding tech improvements:

  • better chat with all the contents and sources,
  • more tools, mostly enterprise ones, for example connection to datalake like databricks,
  • better grid configuration (e.g. moving widgets with drag and drop),
  • speech to text - configure the dashboards with voice,
  • conversation history,
  • more actions and interactions with components (like links, records editing etc.).

Built With

Share this project:

Updates

Submission history