AI Helps Learning

Pam — Sun, 26 Jan 2025 20:12:55 +0000

This is a submission for the Agent.ai Challenge: Assembly of Agents (See Details)

What I Built

AI as an Educational Tool

I am all about learning and creating roadmaps on how to best learn a topic. I used ruby/rails to implement an ai agent and the ruby-openai gem to access openAI's API token. I then created models to represent agents and their interactions as well as creating a controller and service to handle requests and invoke AI agents.
I used Hotwire, turbo for the frontend.

class AiAgentsController < ApplicationController
  def new
  end

  def invoke_agent
    user_query = params[:query]
    begin
      @ai_response = AiService.invoke_agent(user_query)
      Query.create(user_query: user_query, ai_response: @ai_response)

      respond_to do |format|
        format.turbo_stream
        format.html { redirect_to root_path, notice: "Query processed successfully" }
      end
    rescue StandardError => e
      error_message = e.message
      Rails.logger.error("AI Agent Error: #{error_message}")

      respond_to do |format|
        format.turbo_stream { 
          render turbo_stream: turbo_stream.update("response_frame", 
            partial: "error", 
            locals: { message: error_message })
        }
        format.html { redirect_to root_path, alert: error_message }
      end
    end
  end

  def index
    @queries = Query.all
  end
end

require 'openai'

class AiService
  class RateLimitError < StandardError; end
  MAX_RETRIES = 3
  BASE_WAIT_TIME = 10  # Increased base wait time

  def self.invoke_agent(query)
    retries = 0
    begin
      client = OpenAI::Client.new(access_token: ENV['OPENAI_API_KEY'])

      response = client.chat(
        parameters: {
          model: "gpt-4o-mini",
          messages: [
            { role: "system", content: "You are a helpful assistant" },
            { role: "user", content: query }
          ],
          max_tokens: 150,
          temperature: 0.7
        }
      )

      response.dig("choices", 0, "message", "content").to_s.strip
    rescue Faraday::TooManyRequestsError => e
      retries += 1
      if retries <= MAX_RETRIES
        wait_time = BASE_WAIT_TIME * retries  # Linear backoff: 10, 20, 30 seconds
        Rails.logger.warn("Rate limit hit (#{retries}/#{MAX_RETRIES}). Waiting #{wait_time} seconds...")
        sleep(wait_time)
        retry
      else
        Rails.logger.error("Rate limit exceeded. Response: #{e.response&.body}")
        raise RateLimitError, "Please wait a few minutes before trying again"
      end
    rescue StandardError => e
      Rails.logger.error("OpenAI Error: #{e.class} - #{e.message}")
      raise "API Error: #{e.message}"
    end
  end
end

Demo

https://github.com/PamBWillenz/Ai-agent
I used ruby/rails to implement an ai agent.

Home

Query

Query Answers

Agent.ai Experience

Not as intuitive as I thought. I used the platform openai.com. I built my ai agent in ruby/rails and the API key was harder to set up then I thought it would be.
I had to save the key in .env but initiating it was more complex.
The instructions were not clear for the language and framework I was using.

Was the builder tool intuitive and easy to use?
Were the instructions and documentation clear?

No, not exactly. The API key that was generated was in a different format and there were no examples for the framework/language I was using. After many attempts I finally used a curl command to get my framework to connect properly to the AI agent. I used a gem (library in rails) to make the AI api connection.

Labels for any occasion

Pam — Tue, 14 Jan 2025 23:30:36 +0000

This is a submission for the GitHub Copilot Challenge : New Beginnings

I built a label maker in ruby/rails to help automate the laborious task of writing names and addresses from a list to envelopes, if you want to do it the old fashion way and not electronically. Like holidays, thank you's, etc...

Demo -

List Home Page

Holiday Contact View

Labels to Print

My repo -

https://github.com/PamBWillenz/Label-Generator

Copilot Experience -

I used a prompt to get an outline of my project - "I would like to build an app using Ruby, Rails that creates a list of contacts that have names and addresses so I can print labels. Can you give me steps?"
I then asked copilot to recommend gems to use to print labels.
I used chat with the file I wanted to edit, then I would incorporate the edits.
I had an issue with my assets conflicting because I wanted to use bootstrap with simple_form. With Copilot's help I managed to figure out the conflicts and import bootstrap thru node modules vs directly from the gem.
I used 3 different models.
I used autocomplete and also used copilot to generate my tests and my seed file. I have used edits, which I love but did not use that today.

GitHub Models

I started with Claude 3.5 Sonnet, then went to o1 Preview and used GPT 4o the most. When I continually got the same response or the model said I was out of prompts (o1 preview), I switched models. I especially did this when I was troubleshooting.

Conclusion

I feel like New Beginnings mean new ways to approach challenges. My husband and I had a list of more than 200 family and friends that we wanted to send holiday cards, and we had no way to get them in a format to print labels. I wrote all the labels by hand, both for the to: and from:. We can order labels for the from: but you can't do that for the to: so I wanted to automate the process.

I hope this can help others who want to automate their label list printing and use code to automate the process.

Forem: Pam