3 Failures and 1 Win from My Recent Pet Project

Anastasia Karavdina
Aug 10
4 min read

photo by Maria Kostyleva (IG:masha_and_film)

Today I want to tell you how I, with the help of Cursor, built something nobody really needs and why I’m still proud of it.

How it's all started

This summer the number of my mentees dropped quite noticeably. Some had achieved their goals, others decided to take a summer break, and a few had to pause for personal reasons. Naturally, I reacted in the most rational way possible: I panicked. Then I started running free “introductory sessions,” hoping to attract new mentees. That didn’t work at all, because, Anastasia, let’s face it — people who come to chat for free are not usually ready to pay for the kind of service you provide. When will you accept that and stop wasting your time? But these chats did give me something valuable: an idea.

Many people I spoke with asked the same question: “I see Data and AI are growing, and I’d like to switch to that field. Which role would be suitable for me, though?” So I decided to build a product, which will help people to navigate through the variatity of roles in Data field.

Product

I’d heard that nowadays the path from an idea to a working product is surprisingly short, so I decided to approach the building my product in a vibe-coding with You Only Live Once (YOLO) mindset. The concept was simple: a chatbot that suggests possible roles in Data and AI based on a person’s input. Ideally, if the role matched my expertise, it would also subtly recommend me as a mentor and direct them to my website. Armed with this vision, I opened Cursor, got myself a subscription, and started building. The first attempt was me just describing the idea and waiting patiently for magic to happen. The result?

Let’s say… underwhelming. So, I took a more structured approach: I wrote functional and non-functional requirements, listed all the relevant roles along with resources for each, told it to stick to a FastAPI backend with a simple but appealing UI, even chose a color palette. Within a day, I had a pretty solid local version, and in a few more days, I had it deployed to the cloud. That part wasn’t as straightforward as “just deploy to cloud”: I had to specify exactly which services to use. To make it look more professional, I bought a custom domain and ditched the ugly default name.

Launch

When it was ready, I proudly told my friends and two of them tried it. Then I mentioned it in a couple of LinkedIn posts, it brought seven more users. I even created a Facebook ad, which brought two more people. That’s how I reached a grand total of eleven users in the first week. I expected much more, so my impression so far that the world doesn't really need a product like this.

Cost

€17.17 for Cursor, €1.60 for cloud hosting, €0.10 for LLM usage, and €8.50 for ads.

Failures

1. Wasted too much time assuming Cursor could read my mind and deliver exactly what I wanted without me spelling it out

2. My first version was far too simplistic: it didn’t consider things like a user’s experience, skills, or preferences. This led to 90% bounce rate after the first message :(

3. I completely forgot my main goal: to promote my website and my services. I didn’t even include it in the chatbot flow, so unsurprisingly, it brought me exactly zero new mentees.

Win

I love building things but, between you and me, I don’t like coding. I always looked at the coding as obligatory thing to build staff and very happy to outsource large part of it to the machine. So the quick "time-to-market" between idea and released result was an amazing experience.

Next iteration

Since the release, I’ve made a few changes.

The model now runs a more thoughtful analysis before suggesting roles, factoring in whether someone prefers tech or non-tech work, their skills, and their level of experience.
I’ve switched to GPT-5 Nano
Updated the design to a more “terminal-style” look
Added a “Slavic Coach” mode for those who appreciate very direct feedback. The "Slavic coach" is based on the ancient tradition of passing down folk wisdom from generation to generation by repeating proverbs, whether they fit the situation or not, and by speaking bluntly, because life is too short to mince words.
I’ve tried to smooth the edges a bit, but if he happens to offend you, please don’t take it personally: he didn’t mean to hurt you, he’s just never been to a therapist.