Notebook

From thesis to OpenSoft: job hunting with an AI in the loop

How I used Teamlyzer and Claude to research companies and prepare for OpenSoft's three-phase interview, and why the job fits my thesis.

Read in Português

Contents

I finished my MSc in October 2026 and spent the weeks around it looking for my first full-time engineering role. My dissertation argued that AI is most useful when the process around it is explicit: who decides, what gets checked, and what counts as done. I applied the same rule to my job search. Claude did the research and the drafting. I made the decisions.

On 16 November 2026 I start as a Software Engineer at OpenSoft.

The problem

Applying well takes time. A good application needs research on the company, a cover letter that answers the job ad point by point, and preparation for an interview format I usually only half know. When I do that for several companies at once, the quality drops. Letting a model write the letters saves the time and produces generic letters.

What I tried

Teamlyzer first, for every company

Teamlyzer collects reviews of Portuguese tech companies from people who work there and from people who interviewed there. I used it constantly: to decide where to apply at all, to read how a company treats its engineers, and to see what its interviews look like. Before preparing anything, I wanted to know what past candidates had been asked.

One cover letter per role, built from the ad

For each application, I pasted the job description and Claude drafted a letter that mapped each requirement to something I have done: the Gescol HR module at Caixa Mágica, teaching in the AI Agents bootcamp, Apex. My rule was that every claim had to point to a real project. If it couldn’t, the claim went. I edited each draft line by line and wrote the opening myself.

A study plan built from what candidates report

For OpenSoft, I asked Claude to turn the 25 interview reviews on Teamlyzer into a study plan built around what those candidates were asked. The reviews described a friendly process, rated 2.1 out of 5 for difficulty, with a technical part covering Java, OOP, SQL and LeetCode-style problems.

The result was a fourteen-day plan, from 7 to 21 October, of about two to three hours a day. Each day mixed Java theory (a 105-question bank), live coding (25 drills, done timed and thinking aloud) and SQL (joins, GROUP BY, normalisation, schema design on paper first). The last days were two mock interviews, both in Portuguese, because I wanted to practise saying the answers out loud.

The interview, phase by phase

The real process matched what Teamlyzer had described.

  1. Background interview: a conversation to get to know me: my academic path, the projects I had worked on, and what I was looking for.
  2. Technical interview, in two parts:
    • Java: one live-coding problem, plus theory questions on object-oriented programming and the language itself.
    • Databases: I designed the schema for a camera system with three tables: the photo, its metadata and its location, and how they relate. Then came theory questions on databases.
  3. Cultural interview, the conversation I enjoyed most. We ended up talking mostly about my thesis (the defence has its own post), because OpenSoft had been working on the same problem.

Why OpenSoft fits

My dissertation asks how to use AI across the software lifecycle without losing rigour, traceability and quality. OpenSoft deals with that every day. The company is certified under ISO 9001 (quality management) and ISO 27001 (information security), and is appraised at CMMI-DEV v3.0 maturity level 3. In practice, full traceability from requirements to architecture, user stories and code is part of how they deliver software.

That is the chain Apex tries to keep intact when AI enters it. In our conversations after the interviews, they said the work could feed into their own innovation efforts. I am joining a team that already works the way my thesis argues software should be built, and that wants to find out how AI fits into it.

What I would tell someone doing the same

  • Read what candidates say before you study. Teamlyzer told me more about the interview than any job ad. “Prepare me for a Java interview” gets you a textbook. “Prepare me for this company, based on what its candidates report” gets you a plan.
  • Keep your own voice. The drafts were useful because I rewrote them. A recruiter reads many letters a day and can tell which ones were written for them.
  • Practise out loud. Most of the gain came from the mock interviews.
  • Bring your own work into the room. Most of the final interview was about the thesis.