Vertex-Deconstructor

AI assistant that helps users learn the Python version of what they are creating in Vertex AI

Comment 1

🔭 1. Project Overview

  • Concept and Vision: design and implement an assistant for users creating agents in Vertex AI, enriching the learning experience by providing contextual explanations and python code samples, consolidating the user's understanding of the logic and machine learning principles behind their creations.
  • Functionality Overview: we created a tool named Web Search which retrieves technical web content related to the tool, agent, or application mentioned in the user query, providing a breakdown of the technologies and systems involved in the user's creation.

📈 2. Development Journey

  • Initial Inspiration: considering Vertex AI's low-code nature, we wanted to create a tool that let users dive deeper into the programming, logic, and machine learning principles behind the technology that makes Vertex agents work.
  • Design and Development Process: we imagined what we wanted the inputs and outputs of such an agent to look like, and used this as the basis for developing our agent.
  • Challenges Encountered: having not used Google Cloud before, we found navigating a new cloud console challenging, but enjoyed the journey and overcame the learning curve.

🏆 3. Achievements and Learnings

  • Notable Accomplishments: making our first Vertex AI agent, and using our design to facilitate our own understanding of how Vertex AI and Google Cloud work by creating a self-referential use case that refers to the underlying concepts involved.
  • Key Lessons Learned: no-code platforms are exciting, but there is still a significant learning curve involved with setting up APIs and authentication.

🔮 4. Future Directions

  • Upcoming Enhancements: after our first iteration, the main goal is creating a working version of our project. What I imagine we will need to do for this project to work is set up:
    1. A tool for making a google search query based on a user's request, returning a list of urls related to the query
    2. A tool for scraping each of these URLs, putting the HTML into a data store object
    3. A tool or agent for creating a summary of the information out of the data store object These components should map user queries to concise, relevant explainers in chat.
  • Long-term Goals for Vertex-Deconstructor: integrate multiple agents with specific expertise for more detailed and effective agent and tool explainers.

Built With

Share this project:

Updates