Lauri Lappalainen

Job searching can quickly become repetitive and mentally draining. I wanted to make the process more efficient without spending all of my free time searching for roles, preparing applications and organizing information.

At the same time, I had recently received feedback in a recruitment process that more hands-on AI experience would strengthen my profile. Building something useful for my own job search gave me a real problem to solve while developing that experience.

Before building the workflow, I first used AI to understand how agents work, what they are good at and how they should be instructed.

From there, I developed a detailed set of instructions covering the types of roles I was looking for, location and other practical criteria, and how each opportunity should be evaluated and documented.

The first version worked reasonably well, but the workflow developed through continuous iteration. I refined the tone and structure of application texts several times and changed how the information was organized based on how I actually used it.

For example, information about each role was initially stored separately. I later combined the job details, AI-assisted fit assessment and application letter into one document with separate tabs, making the information much easier to access if a recruiter called unexpectedly.

AI works best for me as both a thinking partner and a way to reduce repetitive manual work. The quality of the outcome depends heavily on defining the need clearly, testing the output and being willing to iterate until the result is genuinely useful.

I enjoy customizing user interfaces and tend to pay attention to efficiency, clarity and visual consistency.

Some of the changes I wanted in World of Warcraft required running scripts manually every time the game was launched. One example was a dual-monitor setup where the main gameplay area stayed on one screen while maps, chat and other interface elements were moved to the second.

I used AI first as a sparring partner to understand what needed to be considered and to identify the relevant in-game scripts.

I tested the scripts manually before building anything permanent. Once the solution worked, AI suggested turning it into an add-on that could apply the settings automatically whenever I logged in.

I did not write the add-on code myself. Instead, I described what I wanted, tested the results and used AI to generate and refine the implementation.

I later used the same approach for several smaller interface improvements. Some worked immediately, while others required repeated trial and error. A few ideas eventually reached the technical limits of the game or exposed gaps in how clearly I had described the requirement.

With a sufficiently clear specification, AI-assisted development can make surprisingly capable solutions accessible even without traditional software development skills.

It has also taught me that not every idea is technically possible, and that effective prompting is less about finding a magic phrase and more about understanding the problem, explaining the desired outcome and iterating based on what actually happens.

While working at Foodora, I was the local Finnish stakeholder for customer-service AI features developed by the company’s central teams.

My role was not to build the underlying AI. I was responsible for helping bring the features into real operational use in Finland: communicating changes to the team, acting as the local point of contact, collecting user feedback and development ideas, and providing that feedback back to the central development teams.

One of the first features used the content of a customer conversation to identify whether the contact had entered customer service under the correct contact reason.

Previously, agents sometimes had to correct the classification manually when customers had selected the wrong route in the support flow. AI was introduced to help recognize these situations and reduce unnecessary manual work.

A later feature acted as an AI-supported assistant during customer conversations.

It used the relevant standard operating procedures and what was happening in the conversation to provide the agent with step-by-step guidance on what to do next — for example, what compensation to provide or which action to take on an order.

The purpose was practical: reduce the need to move between the customer conversation and a separate SOP site, speed up handling, reduce mistakes and support more consistent process adherence.

These projects gave me hands-on experience of what it takes to introduce AI into an operational environment: the technology matters, but so do usability, communication, adoption and the feedback loop between users and the people developing the solution.

Create a personal, bilingual portfolio that goes beyond a traditional CV and makes my experience and approach to problem-solving visible.

I use ChatGPT as a thinking partner to challenge ideas, shape the content and turn our decisions into clear instructions for v0. I review the website, decide what I would like to improve, often share screenshots and ask for an assessment before presenting my own preferred solution. Then I test and iterate on the result. My role is to keep the overall direction and decisions in my hands, not to claim I wrote the site's code.

Partway through the project, relying on chat history to recover earlier decisions became difficult. I introduced a shared Google Docs content bank as a single source of truth for approved wording, changes and decisions — a familiar project-management practice from my previous work that proved equally useful in AI-assisted development. It gives me more confidence that earlier and newer decisions won't get mixed up.

The best results don't have to come from one AI tool. Different tools can complement each other, and one AI can help translate a conversation into effective instructions for another. I've also become better at giving instructions myself and discovered that talking through ideas in Finnish can work remarkably well. Most of all, I've enjoyed shaping, refining and publishing something that feels like my own work, even though AI makes the implementation possible.

Tools: ChatGPT, v0, Google Docs.