Productivia

Productivia is an AI agent that helps you optimize your life, by optimizing your activities.

Comment

Inspiration

The inspiration was to build something like existing task systems, such as Google Keep and Todoist, but integrated with an AI agent.

What it does

This is an AI agent that allows the user to interact with a task management system.

How we built it

  1. I built a task management tool with Next.js + PostgreSQL, deployed to a GCP VM instance.
  2. I created an AI agent using the Vertex AI agent builder.
  3. I defined and saved the API for the tool using the OpenAPI spec in YAML.
  4. I tested the AI agent using the built-in conversation UI.

Challenges we ran into

  1. The tool's REST API needed to be a domain with HTTPS. With only an HTTP IP a REJECTED_BAD_REQUEST error was given.
  2. Suggestions from the LLM couldn't be used as input for the tool, a "Sorry something went wrong, can you repeat?" error was given. This error was also seen when copy/pasting in rapid succession, e.g. from a test script, which suggests an internal rate limiting issue.
  3. The tool didn't always work seamlessly or as expected. I suspect that this is due to the use of Gemini 1.0, and that 1.5 would give a more seamless experience.

Accomplishments that we're proud of

A functional agent built with Vertex AI Agent Builder.

What we learned

The Vertex AI Agent Builder makes it much faster to develop agents, but there's room for improvement.

What's next for Productivia

I'd like to see the agent run with Gemini 1.5.

Built With

Share this project:

Updates