Week 3 is now done and pushed my skills beyond what I already knew.
This week was all about combining the tools that I had been using individually to create a workflow for a potential client. I started by looking at various AI tools and their specialities, from ChatGPT and Gemini for text to MidJourney and DALL•E for images, and exploring how they could be combined. ChatGPT, my ever-faithful teacher, gave me a practice activity of having a client (another coffee shop; I think OpenAI is paid to advertise) who wants to create a system for product ideas.
I started by using Gemini (or, rather, ChatGPT impersonating Gemini) to write a product description for a new fall-themed latte the shop wanted to launch. The description was barebones and to-the-point, telling the customer what exactly they’re drinking with little more.
Then, I gave that to ChatGPT to change the tone to something more casual for a social media post. It took the same information but put a spin on it, a personal touch to attract customers.
Finally, it went to DALL•E, ChatGPT’s internal image generator, for an image mockup for a social media announcement post, complete with a detailed background and some leaves on the table.
Gemini Output: “Pumpkin Maple Oat Latte – A cozy blend of espresso, oat milk, and spiced pumpkin, finished with a drizzle of maple sweetness. Comfort in a cup.”
ChatGPT Output: “Fall in a cup. 🍁 Meet our Pumpkin Maple Oat Latte — creamy oat milk, warm spices, and a touch of maple magic. Limited time only!”
This process allowed me to use the different prompting methods that I learned last week as I gave the chatbots examples to base their descriptions off of and created the workflow. After the practice, I started my own assignment that was near and dear to me – a student trying to be a better studier. Since I’m the target client, I tested it on my own notes and classes to see the results firsthand.
I started by defining the input from the student, trying to make it as simple as possible, and deciding to use a page of student-taken notes and the lesson slides. My goal was to use these to create a quick-and-easy review sheet with some practice questions, so that when the material comes back on a test, the student is ready to study.
I started by giving Claude, Anthropic’s AI model, the notes and slides with a prompt to turn them into bullet points. It gave a five-page summary of the lesson (my notes aren’t nearly that long), which wasn’t helpful for studying, but it did clean it all up and format it nicely.
Then, I gave that PDF summary to ChatGPT, this time specifically defining that the output shouldn’t be nearly as extensive as Claude’s. It gave me a one-page summary of the material (much more similar to my notes) and five review questions to quiz myself on. This would be much more useful for an actual student.
Claude Input: “Make a key summary from these notes, using the bullet points as guidelines for what was important but expanding on them with information from the powerpoint”
Raw Claude Output
ChatGPT Input: “Use this PDF to create a study guide for the topic, complete with quiz questions for review. Because this is one lesson and a whole unit, the study guide shouldn’t be extensive, but enough for a simple review session or to be part of a bigger unit of study later on.”
Raw ChatGPT Output
The final step would be NotebookLM, Google’s tool that lets you upload documents, videos, and presentations to use as sources to directly interact with. If the student continued this process for each lesson throughout the semester, uploading ChatGPT’s study pages, NotebookLM would allow them to study more effectively and seamlessly, with everything simplified and stored together.
Overall, this entire process was a lot more streamlined and user-friendly than I thought it would be. If I were to have other clients who required other outputs (images, code, etc.), then more complex tools would be needed, but this gave me a solid foundation to build off of. Once the process was finished, I immediately noticed the high potential for automation that this has, where automation tools can be used succinctly with the chatbots to minimize user input and time required for the workflow. Before turning this completely over to automation, I would have to work to clean up the prompts, making them as consistent enough to reliably get the desired results.
On the business side, this showed me how workflows can scale across different industries and how I can package them as services. It yet again proved how much potential freelancing has, especially if I’m able to create a portfolio’s worth of tools for multiple audiences, all of which are hands-off after the initial input and consistently create deliverable results.
Next week takes this practice and looks at how to turn the workflows into services that I can sell, emphasizing communication with the client.
Final Score: 18.5/20, with the points marked off because I didn’t show my work. Oh, the irony in a chatbot telling me to show my work.
If you want to see my full work, including the original prompts, results, and final essay, I’ve linked the Google Doc here.
Boston Dato 
Leave a comment