Hello, I'm Otsuka, CTO at Liberogic. I've come to realize that building internal apps has become something you can do on a whim. Claude Code was the biggest catalyst for this.
The Evolution of LLMs and Changes in Cost
Over the past few years, LLMs (Large Language Models) have grown in both variety and accuracy, while their cost-performance has steadily improved.
OpenAI recently announced gpt-oss, and it has already been integrated into various LLM hosting services. Looking at the pricing for gpt-oss-120b on Groq, it is remarkably inexpensive.
Even with complex prompts, if usage is on the order of a few hundred runs per month, the cost is negligible.
It Started with "Contact Form Summarization"
Our website has a contact form, and we have long operated a system that uses AI to summarize its contents and send them to Slack.
Make your contact form smarter! Using AI is the best way to handle spam and sales emails.
However, we were using a simple prompt with gpt-4o. We wanted to improve accuracy by using gpt-oss.
"Wouldn't it be interesting if the staff who actually respond to inquiries could write their own prompts?"
However, testing a prompt required running a Node program that calls the LLM API, with past inquiry data pasted in. For engineers, this is easy, but it's a bit of a hurdle for directors.
That led us to the idea of building an internal-only "playground" for AI services.
What we built is a single-page web app.
The specifications are very simple.
- A prompt input field on the left
- Loads past inquiry data saved in Notion
- Calls the Groq API and displays the results on the right
Although it's just this many features, it's an environment where the director can try it out right away.
In the past, even an internal tool of this scale took time to prepare infrastructure and set up servers. Now you can get one up and running quickly on an environment like Cloudflare Workers. Once the requirements are settled, Claude Code can build the program portion in just a few minutes.
Only about two hours were needed
From the moment the idea came to mind until it was finished took just about two hours. *
It was published the same day, and members of the team, including the directors, could try it right away.
What stands out to me is not that non-engineers can now use AI.
"Because internal apps can be built quickly, people around you can be drawn into the development process, not just engineers." This is the most interesting part.
Purpose-built tools can be created in a short time, and the whole team can test them and improve them together.
I feel that this kind of momentum will make upcoming projects much more exciting.
* The most time-consuming part was not the program itself, but understanding how to place the client-side script using Hono + Vite.
The backbone of Liberogic's technical division. Whenever he hears someone say, "I wish we had something like this, it would be so handy," he adds his own ingenuity, turns the idea into added value, and gets it implemented in no time. A highly sociable team member who has many fans among our clients, he is a treasured asset to our company and a devoted cat lover.
Sho Otsuka
Director, CTO / Chief Engineer / Representative, Nekoana LLC / Looks unreasonably young