Agents

Smart Buddy Builder – Your Personalized AI Agent Factory

Comment

Inspiration

Months ago, I was juggling too many things—taking care of kids including a newborn, trying to venture into a business as I lost my job while on leave and learning new AI tools all at once. In a social media group, I noticed several people are sharing their prompts. Those prompts look impressive. But I do not know how to efficiently use and store them. Also, I notice that there are users, who want to monetize their custom GPTs. I was drowning with a lot of tasks and felt so overwhelmed with many AI tools. As for me, as a busy parent, I also need a lot of "assistants" - water reminder, baby recipes, story teller, daily recipes, daily motivations, song writer, image generator. I want an organized way to access my “dream team”.

In my searching and learning process, I ended up as confused and overwhelmed. I believe that there must be an easy way for non-technical people to create and monetize their own AI agents. Surely, for other people, they also have their idea of their "dream team". It led me to create Smart Buddy Builder.

What it does

  1. A user signs in a web portal.
  2. The user is presented with a Dashboard.
  3. The user can create an "AI Buddy" - configuring the prompt for Brain, Heart, Left Hand, Right Hand.
  4. The user can customize the photo. The user can upload PDF documents for added knowledge.
  5. The user can create more "AI Buddy" as long as it is within the limits set by the subscription plan.
  6. In the Dashboard, the user can select an "AI Buddy" and can start to chat with the it. Input can be text and image. Output is currently text.
  7. The user can download the AI buddy. The output is an HTML, that the user can share with others. Optionally, the user can add a paywall to it to monetize the AI Buddy (This is out of scope in this web application)
  8. There is a Setup portal to customize some web application field like default values for configuration options for Brain, Heart, Left Hand and Right Hand.

How we built it

This web application is 99% built using Bolt. I made some manual interventions such as typographical changes to save tokens. I just put my idea by writing a prompt and asking Bolt agent/assistant to do it. I did not create a requirements documents at the start.

Challenges we ran into

  1. Integrations - Supabase RLS and Stripe Webhooks was challenging to integrate. I run into several issues.
  2. Deciding the target audience. Should it be for kids or parents or entrepreneurs?
  3. Tokens running out
  4. How to test LLMs the cost effective way?
  5. The voice input to the chat modal did not work at initial prompt. I ended up removing it for now.
  6. I wanted users to sign-in using Google or github. But remaining time is not enough to test the application with this update. I ended up removing it for now.
  7. I need to check more about security issues.
  8. I need to think about legal considerations of this web application before making it public (ex. Is there concern about data security and privacy?)
  9. I need to think about trademark and rules in my country before making this as a business.
  10. I am unsure how the usage impacts the operating costs of this web application. Ex. Is the pricing and usage limits fair? Should I just open source the code?
  11. How to use the tokens effectively, especially if the code size is already big? I was getting a warning about large size of the project. But I do not know exactly how I can effectively use the tokens. I tried to follow the suggestion of duplicating the project. Still, I got same warning about large size of the project.
  12. I redesigned the landing page on the last day of the hackathon. I should have been clear from the start.
  13. I have worked as a software tester. I know the traditional way to release a software. But in this hackathon, I needed to release even if I have not fully tested all features (ex. edge cases)
  14. I wanted a lot more features. But I need to finish something by the deadline. (Ex. be able to choose what LLM each agent will use)

Accomplishments that we're proud of

This web application is actually working. I can build my own AI Team for my own use.

What we learned

  1. I can make a web application using prompts.
  2. It could be better if I documented a Software Requirements Specification at first before I started putting the prompt to Bolt. Or I should be clear from the start what my MVP is.

What's next for Agents

  1. Build the ‘Canva' of AI agents.
  2. Agent community, embeddable widgets, AI-on-AI (buddies that train each other)
  3. Platform and Tools integrations
  4. Multi-modal input and output. Ex. Use of voice input and output
  5. Use of MCP and Agentic AI -> will that be possible

Built With

Share this project:

Updates