Oliver Kumpulainen

Studying AI. Building tools for real work.

I’m an Information Technology student at the University of Oulu, specialising in artificial intelligence. I’m the founder of Werwit, and I build my own software on the side: a training app that plans my next workout, and a research pipeline for testing gold trading strategies.

  • Oulu, Finland
  • Available for summer 2027
  • AI & software development
Oliver Kumpulainen

What I’m building now.

Own projects, built alongside my studies. Each one starts from a problem I have myself.

An outfit look from Werwit's demo content AI image
iOS · in testingStartup · Founder

Werwit

Werwit turns a picture into an outfit from your own closet and what you can buy in Finland. I started it because I didn’t know what to wear, and didn’t want to open ten online stores to find one pair of trousers.

My role
Founder: product, brand, and building the app and the website
Where
Startup Oulu AI incubator
werwit.app
Home screen: morning check-in, readiness score and the next workout Workout screen: warm-up sets, top set and backoff sets with RIR targets History screen, including workouts imported from old notes
In developmentPersonal project · iOS

A gym app that plans the next workout

My lifts stalled because I couldn’t track progression. So I built a native iOS app where a new user sets up their own program and exercises, and the app learns from every logged set to give the next workout with progression built in.

How it works
A morning check-in (sleep, energy, soreness) gives a readiness score for the day. Each lift has a top set and backoff sets with RIR targets, and “Why these weights?” explains the numbers.
Details
Swap an exercise when the machine is taken. Notes for machine settings. My old workouts imported from notes.
Stack
SwiftUI, Supabase
SignalAudit report: headline figures with costs included, and a count of every signal by status SignalAudit · every signal accounted for
In developmentPersonal project

Gold strategy research pipeline & SignalAudit

A backtesting and validation system for gold trading strategies, run by a team of Claude Code agents (researcher, developer, risk reviewer, reporter) under strict rules: separate development, validation and holdout periods, no look-ahead, trading costs always included, every experiment logged. It never trades real money.

The second part, SignalAudit, replays the signals that Telegram “signal groups” publish against real minute-level gold prices with spread, commission and swap, and shows what actually happened, including the signals that never filled or can’t be resolved.

Stack
Python (pandas, NumPy, pytest, Hypothesis), yfinance / HistData, FastAPI, SQLite, Next.js, TypeScript, Tailwind, Playwright, Docker Compose, Claude Code subagents
What I learned
How easy it is to fool yourself in a backtest, and what discipline prevents it: pre-registration, holdout data, independent review. How to design multi-agent workflows where agents challenge each other’s results instead of agreeing. And that most well-known trading rules don’t survive realistic costs.

Before that: industry.

Summer Trainee at Valmet Flow Control, Hakkila, 2026. On call since. I started in logistics and ended up building tools for the people around me.

The app's main menu: paint shop registers a new actuator, shelving adds a shelf location, picking finds an actuator Main menu · the three steps of the flow
In useValmet Flow Control

Warehouse management app

Items were tracked on paper, so finding things took time for the logistics staff and their supervisor. When the team asked if I could build them an app, I built it from start to finish. It is still in use.

The flow
The paint shop registers a new actuator, shelving gives it a shelf location, and picking finds it. Three buttons, one for each team.
Stack
Microsoft Power Apps, developed with AI assistance and MCP
Why Power Apps
It fit straight into the company’s Microsoft environment, so it was easy to deploy
Machinery visualisation I made
Finished LOTO safety sign, made as a team
In useTeam project · my focus: visualisations

Machinery visualisations for a LOTO project

LOTO (lockout/tagout) keeps machines isolated from their energy sources during servicing. I produced the machinery visualisations with AI image generation, and with my colleagues mapped the energy isolation points and made the safety signs.

Prototype · paused

Product code parser

Type-plate codes were decoded by searching PDF documents. I built a working Power Apps prototype that splits a code into its meaningful parts. It works in my own environment; development is paused and it hasn’t been deployed.

Microsoft Power Apps

What I’m learning.

Not finished work. This is where I’m spending my evenings, and what I want to do more of in a team.

AI engineering

  • Building with Claude Code: skills, hooks, subagents
  • Model Context Protocol (MCP) servers and tool integrations
  • Multi-agent workflows where agents review each other, used in practice in the gold research pipeline
  • Context engineering

About

I like taking an everyday problem and turning it into a tool someone can actually use.

I’ve studied Information Technology at the University of Oulu since 2024, on the B.Sc. and M.Sc. (Tech.) track, focusing on artificial intelligence. My studies cover AI, software engineering, digital image processing and business data analytics, and I’ve taken courses in industrial engineering and management. I’m most interested in where software, industry and business meet.

Before civilian studies I spent a year (2023–⁠24) in the Air Force programme at the Finnish National Defence University. I completed Reserve Officer Course 261 and later served as an instructor on course 262, where I learned to take responsibility, lead groups and teach others.

Languages
Python, Java, TypeScript, Swift
Tools
React, Supabase, SwiftUI, Microsoft Power Apps, Git, Claude Code, MCP
  1. 2026 →WerwitFounder · Startup Oulu AI incubator
  2. 2026 →Valmet Flow ControlSummer Trainee, then on call
  3. 2025Lidl Finland, Laukaa DCLogistics
  4. 2024 →University of OuluInformation Technology, AI
  5. 2023–24National Defence UniversityAir Force programme · RUK 261, instructor on 262
  6. 2022Laatulavat Oyvia Link Henkilöstöpalvelut
  7. 2021PostiEarly-morning delivery

Let’s build something useful.

I’m looking for AI and software development roles where I can work on real problems, and I’m available for summer 2027.