RoboGenie

• 

  Landing page

• 

  Creating a new job

• 

  Answering the AI's questions about the job

• 

  The job is running

• 

  Result of running the job

Inspiration

I often find myself building simple web scrapers which send me updates via email. But most non-hackers will never even try doing this. With RoboGenie, they can create notification conditions in plain english!

What it does

RoboGenie can notify you about anything. For example:

“Monitor the weather forecast and when it's going to be below 20F, send me an email with the prices to fly to Cancun”

How we built it

When you make a new request, an LLM agent makes a plan and asks you clarification questions. Then, the task is scheduled to execute at a specific time.

Upon execution, a new agent runs which has access to a web browser and the ability to send you an email. It can also take persistent notes on a scratchpad and reschedule itself to run at a later time, enabling infinite flexibility with ongoing tasks.

Challenges we ran into

Getting the LLM to understand the content of a web-page and how to interact with it is quite challenging. I tried visual models like GPT-4o, and I tried feeding the entire DOM to the LLM, before eventually building a custom DOM-to-LLM-readable-tree conversion process.

Accomplishments that we're proud of

This is probably the smoothest product experience I've ever built at a hackathon.

What we learned

I learned how to build more complicated agent workflows and what works and doesn't work in terms of LLM function calling. For example, Claude struggles with deeply-nested input especially when there are many tools available.

What's next for RoboGenie

It's pretty much production-ready so I plan to promote it on socials and see what happens!

Built With

• next.js
• react
• typescript