Fullstack developer Vladislav Harkusha shared a case study of AI integration for an online store and talked about the difficulties that drove him «a little crazy.»
Fullstack developer Vladislav Harkusha shared a case study of AI integration for an online store and talked about the difficulties that drove him «a little crazy.»
IT specialist had long planned to create an intelligent chatbot for an online shoe store. At the initial stage, the process went smoothly: Vladislav registered on the n8n automation platform, prepared the necessary prompt, and configured the basic logic. Then, he created an intermediate server on the Next.js framework for direct access to n8n, where GPT-5mini was connected as the main language model.
However, during the testing process, the developer encountered a problem familiar to many creators of AI-based solutions.
«I almost went crazy. At first everything was fine, but then what many vibecoders talk about started happening — tokens! This whole system started eating a lot of tokens,» Vladislav shares on LinkedIn.
To optimize resource consumption, the developer initially implemented a limit of 6 messages per 30 minutes. However, a front-end solution alone proved insufficient for full protection and control.
«On the front end, this wasn’t enough, so I had to do everything „like a human“: I made a router that contacts n8n and returns the products I need, and also added a limit on the backend. Then I reduced the context and shortened the response to only what was really needed,» explains the IT specialist.
According to Vladislav, the task, which was initially planned to be completed in approximately two hours, ended up taking half a day of intensive work.
Despite the difficulties and significant timing adjustments, the developer evaluates this case positively, because thanks to architecture optimization, he managed to gain valuable practical experience and achieve the desired result.
