Ideate

bring your ideas to life

Comment

Inspiration

You know when you're with a friend and you're trying to explain something but it gets a little too serious so you have to pull out the paper, and then you guys exchange the single pencil you have? We want to emulate that experience with AI chatbots, basically. We want to give people the power to push themselves beyond their imagination, hence why we picked ideate. Beyond whiteboard, we give users a markdown based text editor and an ide to test code with. Whether its devs coming up with system designs, business owners sketching out their business plans or people just trying to have fun, we believe Ideate is a great promoter of creativity and productivity.

What it does

  • Ideate is a fully encapsulated Whiteboarding tool, Python IDE, and Markdown editor with a deeply engrained AI agent with a perfectly engineered system prompt and context control. It allows for the entire software development suite to be done in a single page, from design to creation and testing to documentation.

How we built it

Google Gemini + Excalidraw

Challenges we ran into

Coming up with the optimal approach to making the most out of AI in Ideate. We tried out different approaches from summarizers, to planners to routers. In the end, we settled on a router to planner approach that used lightweight models. This allowed us to push the accuracy of our system without having to sacrifice too much time. Managing context of the user’s workspace was also a challenge. We needed just the right information but also not too much as we tried to balance inaccuracies and latency.

Accomplishments that we're proud of

  • Creating a responsive, fully fleshed out front-end that allowed for modularity while also looking nice
  • Implementing an extremely lightweight python editor and interpreter that could have code programmatically injected into it by an llm.
  • Routing requests to different handlers based on data type
  • Creating a unique interface where a person and an LLM can collaborate in real time through visuals
  • Designing a system of response streaming in which doodle buddy follows a to-do list, updating the user as soon as tasks are completed (as opposed to forcing the user to wait for a long time)

What we learned

  • Using Mastra agents to handle complex tasks
  • Prompt engineering and formatting to optimize LLM performance
  • Connecting our React frontend with our AI backend

What's next for Ideate

Voice to voice interactions while whiteboard. Completely interactive experience. Live collaboration with other people and multiple AI agents more modules and tools to work with for both humans and agents Further optimize LLM speed and reliability

Video Links

https://www.loom.com/share/f55e92112e1d4574a412de91353d0190?sid=8ef4c78c-3651-41bd-b544-7eeae484d08f https://www.loom.com/share/1e839077c51445f08c3dd8a0ab05dd1f?sid=02a2d729-3e01-4597-ae39-9c29f36b605a

Built With

Share this project:

Updates