I tried making a 'bot that can answer anything about childcare' with Gemini

"Ah, so engineers are geniuses after all."
I felt that strongly after trying my hand at pseudo-programming using Google's generative AI, Gemini.

I'm the type of person who eats their favorite food first. I want to savor the happiness of eating the most delicious thing when my appetite is at its best. On the other hand, my wife clears away the things she dislikes first and saves the reward for last.

This wasn't just true for meals, but also for childbirth preparations. Before I knew it, my wife had acquired various knowledge and gathered the necessary items. She was steadily preparing both her surroundings and her mentality for the birth.

Meanwhile, with the birth of our first child just a week away, I suddenly started feeling anxious. Although I had looked through the how-to books on basic childcare knowledge that my wife had bought, I wasn't confident that I could put them into practice immediately. It was because I had only been eating the things I liked and putting off the things I needed to do or learn.

"They say that for the first month after giving birth, a new mother is like someone who has suffered a serious injury."
"I'm not your parent."
"Don't ask me about every little thing."

The words my wife said while laughing flash back to me. That's right, I'm the one who will be feeding the baby, burping them, changing diapers, moisturizing their skin, giving them baths, and putting them to sleep.

If things go on like this, it's going to be bad. I can clearly picture myself in my mind, feeling anxious and flustered, wondering if I'm doing things correctly. Even if I try reading the books again, it doesn't feel like the information is sinking in.

I have no choice but to learn by doing, but this isn't a situation where I can afford trial and error. And I should assume that for at least the first month, there won't be anyone by my side to answer my helpless questions.

If there's no one there, I have to make someone. What if I made a "bot that can answer anything about childcare" and asked it one by one how to do things, whether my methods were correct, and what I should be careful about?

My first programming experience

When I decided to try making a "bot that can answer anything about childcare," generative AI was on my mind. Since I had used it at work, I vaguely thought that if I used AI, I could probably make something like that.

So, I told Google's generative AI, Gemini, what I wanted to do. In response, it gave me a powerful "I can do that" and a well-organized set of steps for creating it.

Alright, so it really is easy after all. As I thought that and read through the results carefully, I realized that the bot I was about to create wouldn't be something that just runs on Gemini. It took me three readings of the output to understand that I would be creating a type of application that runs on the web by linking various tools together.

To put the mechanism of the "bot that can answer anything about childcare" proposed by Gemini simply, it's like this.

If it were just this much, it seems like something I could make with just Gemini, but apparently, that's not how it works. The reason I obediently followed instructions was that when I saw the way the application was explained, I thought it looked kind of interesting. Of course, I had never developed an app, and I had no programming experience. I hadn't even been interested in it. But precisely because it was a world I had never touched, I wanted to take a peek.

Here are the steps for creating the application that Gemini taught me. Using a house as an analogy, it seems like I'll be writing the blueprints using a programming language called "Python" and building it using a tool called "Streamlit."

I stepped into an unknown world with excitement.

I have absolutely no idea where I am.

The first thing I did was register for something called "Google Cloud Platform." It said, "Sign up for a free trial ($300 credit, valid for 90 days)," and I almost decided to stop before I even started. Services that display signs saying "free for XX days" often automatically switch to a paid plan once the period ends, and I have never once managed to cancel a subscription before the deadline.

Feeling anxious, I asked Gemini, and it seems that the paid plan doesn't automatically start after 90 days. Okay, I'll trust it. Registration was surprisingly easy. In the top left of the screen, it says, "¥43,457 credit, 90 days remaining." A feeling that I can no longer turn back from the path I've taken is washing over me. No, it's okay, let's move forward.

The next instruction is to "create a project in the GCP console." Unfamiliar words keep appearing one after another, and I'm starting to feel like I'm lost. In fact, I didn't know where any buttons were on the site, and I was wandering around aimlessly. I completely lost track of where I was and what I was doing. I didn't realize that the GCP in GCP console stood for Google Cloud Platform until after I finished developing the app.

I would complete one instruction, ask Gemini a question, and then go back to work. I chipped away at tasks with names that looked like spells, such as "enabling APIs" and "obtaining JSON keys," one by one.

A capable secretary who can do anything.

The next thing I did was collect and organize the knowledge that would become the core of the "bot that can answer anything about childcare." I created a database of information related to childcare, from how to make formula and how to do skincare, to what to do if a child accidentally swallows something. The idea is that when I ask a question, the AI will extract the appropriate information from that database and give it to me.

Gemini is remarkably helpful here as well. Using a feature called "Deep Research," it brilliantly collected a massive amount of information that would have taken me about an hour to gather, in just about 3 minutes.

And the next step is creating the "blueprint," which is the most important part of this application creation. The task of writing code using a programming language called Python. Of course, there's no way I could write code. However, Gemini can. I even had Gemini write the instructions for getting the code written, and when I pasted them, a feature called "Canvas" for creating and editing code automatically launched and it was written in an instant. I just watched the screen, and the blueprint was completed quite easily.

An update after 10 years.

Finally, I'm starting the work of assembling the application using the various materials I've prepared. Even though I say "finally," it's still the same day I started working. All things considered, it's going smoothly.

I'm using a tool called VS Code for the work. Of course, I've never heard of it, but I'm sure I'll manage if I have Gemini. It seems to be overseas software, so I stared at the English site and proceeded with the download. When I finished and double-clicked to open it, a warning appeared saying, "VS Code cannot be used on this computer's OS."

That's no surprise, as the MacBook Air I'm using was bought for me by my grandmother as a gift for entering university. When I looked it up, the OS was one that came out 10 years ago. Counting from the latest one, it's surprisingly 9 generations old, which seems to be equivalent to an "iPhone 6" if we're talking about iPhones. It's a wonder the computer was even working properly in the first place.

Since it was "impossible!" to update to the latest OS all at once, I gradually evolved it through OSs with hurricane-like names: "El Capitan (2015)" -> "Catalina (2019)" -> "Monterey (2021)." I felt better seeing the computer's home screen look clearly more stylish, and I resumed my work.

Error after error

The tool called VS Code really feels like something made specifically for programming. Just looking at it makes me feel overwhelmed. First, I follow the instructions to create a folder and store the necessary files I've prepared so far. As usual, I don't even know where the buttons or tabs are, so just that alone is a struggle.

Next, I paste in some code. I have no idea what is written there, but since Gemini creates the code itself, I just copy and paste it. At first, I didn't understand what was happening on the screen because I didn't know that after pasting the code and pressing the Enter key, blue or white text meant a normal state, while red text meant an error had occurred.

Even so, I pushed through the thicket and finally reached the stage of running the application. It was late at night. I had been glued to the computer for so long that my lower back started to hurt. But that was almost over. I pasted the generated code and pressed the Enter key.

The screen switched, and a new browser tab opened automatically. In the top right, there was an icon that looked like a stick figure running, along with the text "Running...". Alright, it worked. The moment I started to say that, a massive amount of red text appeared on the screen. I could tell at a glance that an error had occurred. Well, I guess it wouldn't work on the first try. But from here on, it was a long road.

“Oh, so that’s what it was.”

When I copy the error message and ask Gemini, it gives me precise solutions.

However, once I clear one error, another one pops up. With the countless repetitions, the finish line that I had almost seen kept moving further and further away. When I carefully unraveled it, I realized the problem was that a tool called "Streamlit" wasn't working properly.

To use an analogy, it was like trying to build a house where the blueprints (the program code) were finished, but the lumber and tools (the parts and settings needed to run the app) didn't match the blueprints or were outdated, so it couldn't be assembled correctly. Every time I pulled information from the internet or processed it, problems kept popping up one after another, like "This pillar is the wrong size!" or "This nail is rusty and can't be used!"

Completion, and then a realization

While feeling like I was going crazy, I continued the same kind of work endlessly. I was just copying and pasting code as instructed, but as I watched long strings of text appear and disappear without red error messages showing up on VS Code, it gradually started to become fun. I began to understand what I needed to do next, and my fingers started moving on their own.

And then, the application screen launches. Looking at the screen I don't know how many times I've seen, there are no errors. Trembling, I press the microphone button and ask.

“How do I give a baby a bath?”

Then, a reply came back in an unusually cheerful male voice. The same content was also displayed in text. Wow, finally. I actually did it.

“It’s still inconvenient, after all.”

The next morning, I thought to myself while looking at the screen. It’s an app I finished with such great effort, but it can only be used on a computer. During the process, I had consulted Gemini about wanting to make it work on a smartphone, but I was persuaded by being told, "Let's aim to complete a working application first. We can talk about that later," and I was convinced at the time. But in actual childcare situations, when you want to ask something while doing something else, it is clear as day that it would be more convenient if it could be used on a smartphone.

I wondered if there was any way to transfer the functions of the app I created this time so that I could use it as-is on my smartphone. When I asked, it cheerfully threw back, 'In that case, you can do it with a "Gem"!'

'Gem' is a feature that allows you to customize Gemini for specific tasks or purposes. When I think about it, Gemini is equipped with voice input and output functions.

Wishing, 'P-please stop,' I followed Gemini's instructions and started creating a Gem equipped with the functions of a 'bot that can answer anything about childcare.' It was completed in the blink of an eye, surprisingly smoothly and without any errors. Whaaaat.

When I ask a question by voice, a cheerful male voice gives an accurate reply. And, I can also use it on my smartphone. It was a complete, superior version of the application I had worked so hard to create.

By the way, the Gem's database was created using a tool called NotebookLM. Since I hadn't really done anything, I decided that at least the knowledge information to be embedded should be something I had carefully selected myself. NotebookLM is a tool developed by Google, and while generative AI like Gemini generates answers based on general knowledge, it generates answers based on uploaded materials. This means the basis of the information is clear.

I put links to websites and YouTube videos that I thought 'this is it' into NotebookLM, had it summarize the content, and uploaded that to the Gem. Rather than just the summarized text as-is, if you ask Gemini to 'summarize it in YAML format,' it becomes easier for the AI to read.

It also becomes my own personal radio.

Since Gems can input and output via voice, I can talk to it when my hands are full, and it will respond with its voice. If I say, 'Keep talking based on the knowledge for 30 minutes,' it will talk just like a radio.

'Childcare Agent No. 2 (the name of the Gem I made) is here to bring you Infinite Childcare Radio, starting now,' it says, strangely enthusiastic. I go for a 30-minute run late at night, so it's perfect. Even if you find reading books troublesome or have trouble getting information to stick in your head, this allows for forced input.

In conclusion

Since what I wanted to do was realized so easily with Gemini's Gems, the application I had desperately created turned out to be a complete waste of effort.

But, for the first time in a while, it was a lot of fun. I feel like I understand, just a little bit, the feelings of middle and high school students who you sometimes see saying, 'I developed an app by teaching myself.' Absorbing knowledge I didn't know, the moment I could get out of a maze I couldn't escape from—it was just fun and I became obsessed.

By stepping into the world of programming, led by a guide named Gemini, I was able to remember things I had forgotten as an adult. The feeling of thinking something is fun the moment it finally goes well, after it didn't go well, and didn't go well, and didn't go well.

My conversation with Gemini reached about 100,000 characters. That's about the length of a paperback book. It didn't tell me the most important thing, 'You can do that with a Gem,' at the beginning, but it was an incredibly fun and meaningful detour.