Fabio Hase
Ten months on the process side. Twelve on the engineering side. Joining them is the point.
- Modelled and documented the processes of a manufacturer of sugar-beet harvesters, over 400 people, and analysed three years of its production
- Shipped a Flutter app to both app stores and ran it for twelve months; product and engineering were mine
- Took one site manager’s manual process apart in a working session and rebuilt it as a running five-stage pipeline
- B.Eng. Engineering and Management, taught entirely in English. 1.5 after the fourth semester, on the German scale where 1.0 is the top mark
Available January to June 2027 — 20 weeks, for an internship in AI consulting or product.
The work I am looking for: process discovery with the client · deciding which steps are worth automating · building the prototype that proves it · measuring whether it gets used and what it saves.
EU citizen, no work permit needed. English at C1. In Barcelona until the placement starts — Erasmus at IQS, Universitat Ramon Llull.
HOLMER — ten months documenting how a manufacturer actually works
Working student · March–December 2025 · process modelling · quality KPIs
HOLMER builds self-propelled sugar-beet harvesters. Over 400 people work across the group. I modelled and documented the company’s processes, and built the integrated management system. I worked out the KPIs quality control steers by, and prepared the data behind them. I documented the value-stream analysis, and analysed three years of Terra Dos production.
None of that was written at a desk. Every model started with the people who run the process, in the place they run it. What a process owner describes, what the documentation says and what happens on the floor are three different things. You only find the third by standing there.
I also researched the software options for a management handbook. That is a small version of the job I am applying for: read how the work actually runs, then say what should carry it. It is also the only work on this page I did inside an organisation I did not start. That is where you learn what a process looks like when other people have to run it every day.
Seeing where those three come apart is half of the job I am applying for. The other half is building the thing that closes the gap. That is what the twelve months below were.
LQΛL — bikepark discovery for the DACH region
12 months · 2 founders · iOS and Android · shut down 3 August 2026
Four facts decide a bikepark trip: is it open, what does it cost, which trails are there, what is the weather at the summit. They sit scattered across hundreds of separate websites, as PDFs, as images, as trail lists years out of date. The bet was that putting them in one place would make riders use it, and make bikeparks pay to appear there correctly.
I owned product, engineering, strategy and project management. That meant a Flutter app in both stores, a Next.js self-service admin for the twelve partner parks, and a Supabase backend behind both. The backend included an AI assistant answering free-text questions against the park data. It ran server-side, so the key never shipped to a phone. My co-founder owned marketing, sales, the twelve partnerships and the commercial side.
We never tested either half. The numbers answered the bet instead.
142 installs, 26 accounts, €0 revenue. Usage peaked in the closed beta, before the app was ever public, and 81% of the people who opened it never came back for a second day.
I ended it in two steps, and both were mine. In May 2026 I stopped development. Without working user acquisition, every further hour would have gone into a product nobody finds. The first summer was the test: bikepark season, the app live on both stores, eight active users in June and seven in July. At the end of July I closed it formally.
The part I got wrong is not in those numbers. I could see the sales side was not moving, and I spent months compensating by building more. A commit feels like progress; an unanswered email to a bikepark feels like nothing.
We also shut down on the data rather than the bank balance. Hours and prices move with the season, and without the time to keep them right the app would have gone on shipping, only increasingly wrong. The twelve parks had refused to maintain their own data because they had no capacity. We closed for exactly the same reason.
A product built on continuously maintained data is not a build. It is a permanent obligation, and it belongs in the plan before the first line of code.
→ The case study: the funnel from install to registered account, and where the money went
LV-Pilot — from a construction spec to subcontractor prices, compared
One-weekend case study · two of us · 16–17 May 2026
This is the two halves in one build. The process was legible to me because documenting processes is what I had spent ten months doing at HOLMER. Building it was the part I already knew how to do.
Nothing here was invented at a desk. A site manager I know sat down with us in May 2026, and we went through his working day until one process was left. It had to be narrow enough for a weekend and broad enough to be worth building.
Every enquiry for a new build arrives with a Bill of Quantities attached. He works out which trades the job needs, writes to the subcontractors for each one separately, then collects and compares the answers by hand. Several hours per project, by his own account. One thing he said decided everything after it: subcontractors answer by email, and the trade accepts no other tool.
LV-Pilot reads the document and extracts every position. It matches the trades to subcontractors on file and drafts one request per pairing. The site manager confirms them, they go out as ordinary email, and the replies come back classified, so the answers for one trade sit side by side.
Five stages, one weekend, one real Bill of Quantities and fictional subcontractors. The whole chain has run once end to end — four requests out, the first reply classified 81 seconds later — and never inside the company it was designed for.
→ The pipeline: five stages, and the constraint behind each one
- Start with the person who does the work, in the place they do it. The one sentence that decided LV-Pilot’s architecture came from the site manager, not from us
- Decide between options on stated criteria, and write down the one you drop. That session produced two ideas, and the reason the second was dropped is logged beside the one we built
- Where a model is unsure, the person accountable decides. LV-Pilot’s drafts wait for the site manager, and a reply the classifier cannot place goes to him with the mail attached
- Name the test before running it, then let it answer. I stopped development on LQΛL in May and gave the decision to one bikepark season
Three sectors so far: agricultural machinery, outdoor leisure, construction. In each one the facts a decision needed already existed, in a document nobody could work with. Taking that apart is the work I keep coming back to.
Both projects were built next to full-time study. One shipped to two app stores and ran for twelve months. The other went from a working session with a site manager to a running pipeline over a weekend.
They also share a pattern. Twelve bikeparks would not maintain their own data. The tool built for a construction company has not been run there either. Correct work that nobody adopted, twice.
The correction I took from the first one is to sell before building. Ten serious conversations with bikeparks before the first feature would have surfaced the capacity problem while the model was still cheap to change.
That is the part I want to get better at. It is also the honest reason consulting interests me rather than a pure engineering role. I want to be in the room where someone decides whether the work is worth continuing, and to have the numbers to answer it. I can build the thing. What I want next is learning to make it land.
- Build
- Specifying, directing and debugging AI-assisted development, and deciding what a system should do. Shipped in Flutter and Dart with Riverpod, and in TypeScript and Next.js on Supabase, with Postgres data modelling and edge functions. Server-side LLM integration with Claude and OpenAI, PostHog funnels, transactional email and inbound webhooks with Resend. Currently learning: workflow automation with n8n and retrieval-augmented generation (RAG).
- Process
- Process modelling and documentation · value-stream analysis · quality KPI definition, and the data behind it · integrated management systems · production data analysis over three years · requirements taken from the people who do the work
- B.Eng. Engineering and Management, Technische Hochschule Ingolstadt — taught entirely in English
- 1.5 after the fourth semester. German scale, where 1.0 is the top mark
- My degree includes a mandatory 20-week practical semester in an international environment, which is what this application is
- German citizen, so an EU citizen: no work permit and no sponsorship needed anywhere in the EU
- German is my first language; I work in English at C1, certified by the DAAD in April 2026, after a school term in Birmingham, Alabama