Inspiration
This project was inspired by my own frustrating experiences being onboarded into different companies. In general, people provide you link to bunch of documents and set up calls to share their knowledge, you shadow them. It's kind of random and generic, depending on the person onboarding you either your experience can be helpful or frustrating. On the other hand for the team onboarding the new joiner, they kind of just wing it hoping the new joiner will pick up on their own.
So, OnboardingBuddy is an agentic system that develops customized onboarding plan for new joiners and guide them throughout their onboarding period.
What it does
It acts as an onboarding buddy like the name says. It creates the plan and guides the new joiner throughout the process, plus answer all their questions. Additionally, it also gives a view to the manager on the progress of new joiner.
How we built it
It was built using GCP resources, mainly:
- Gemini for planning the app and engineering the prompt for Antigravity for the 1st version
- Antigravity for further iteration to the app
- Cloud run for hosting the app on cloud and provide a public URL
- Cloud storage for acting as knowledge base hosting all the documents that feeds the agents
Challenges we ran into
I found this challenge 2 days before the deadline and decided to submit this project. I had previously tried building this project with claude and as a challenge with this hackathon decided to rebuild it using only Google resources. So everything in this project has been made using Google resources. This was also my first time using Google resources to code and build something and I'm not a developer.
Accomplishments that we're proud of
Built my first end-to-end app using only GCP resources in 2 days.
What we learned
I learned about cloud run, I didn't know about it before this project. I also never used Gemini for code building purpose. Before this project I only used it for general questions, as a replacement for ChatGPT. Claude is what I used for assistance with any technical project. Now I'm aware about the capability of Gemini and also AntiGravity.
What's next for OnboardingBuddy
The UI can be better with a more minimalistic design. Right now there is too much text. On the other hand, the knowledge base is filled with documents that was generated using Gemini. It's pretty much organized but in real-world information is more scattered. So I'd like to experiment with providing it messy information in the knowledge base and use classical ML models to clusters and organize it so that other agents can find information easily. This is just an idea at this point but something that I'd like to try as a next level.
Log in or sign up for Devpost to join the conversation.