Be curious.
The technology changes; the method doesn't.
Projects
- 2026Personal · In progress
The personal AI system I am building for myself, and the base my other projects run on.
Every side project I started needed the same few things: access to models without scattering keys across repositories, a way to search my own knowledge and get answers that show where they came from, and work that keeps running in the background. Rather than rebuild them every time, I am building them once, on a small always-on server, as a system my projects plug into.
The first version was deliberately learning-first: I built each piece by hand to understand how it fits. The second keeps what that taught me and leans on mature tools for the rest. Both are public, decisions and failures included.
- 2026Personal · In progress
A private search assistant over 13 medical textbooks, built for a friend studying medicine.
A friend of mine studying medicine has to work across 13 textbooks, more than 12,500 pages, and does not trust ChatGPT to stay grounded in them. A fluent answer is no use to him if he cannot see where it came from.
So I am building him a private retrieval app: it finds the original passages, shows each one with its book and page, and keeps the short generated answer visibly separate from the sources. It runs on HomeLab and is built in public; if it turns out good enough, it may open up to others.
- 2026Work · Microsoft
Six tutorial videos on manufacturing in Business Central, and the AI-built toolchain behind them.
My first assignment at Microsoft was a six-part video series on the foundations of manufacturing in Business Central, from production BOMs and routings to cost control. Instead of treating it as a one-off, I built a production process around it, and built all of its tooling with GenAI.
Research was grounded in Microsoft Learn through its MCP server, demos were made repeatable with Page Scripting, Copilot agents handled the production steps and an FFmpeg workflow the review. For clean takes I wrote ClickReplay, a small public tool that records and replays mouse and keyboard actions.
I then handed the whole workflow to a content specialist without a coding background, with a setup guide, a process guide and a guided Copilot coach, so the series can continue without me.
- 2026Work · Microsoft
Research into what an AI agent needs before it can act reliably inside an ERP.
Alongside the videos, I opened three connected research workstreams on my own initiative, each prompted by a gap the previous one exposed. The first looked at how an agent can operate Business Central at all, and showed that better interaction mechanics do not give it business understanding. The second looked at how it gets the right context before acting: the product's structure, its documentation, and the setup and live state of the company it works in. The third was an evaluation harness for comparing whole AI pipelines on outcome quality, speed, usage and cost.
It was research, not product: prototypes, evidence and a recommendation, documented and handed over when my contract ended. Nothing shipped, and the comparative evaluation runs were still in progress when I left.
- 2025 – 2026Academic · CBS MSc thesis
Does quantum computing give finance a practical advantage yet? Not yet, and the bottleneck is not the algorithms.
My master's thesis, co-authored with Vincent Wallerich, asked whether gate-based quantum computing delivers practical advantage in finance. We screened 6,232 records down to 777 papers and synthesised them across eight problem areas, with an AI-supported thematic-analysis workflow that used several LLMs for coding, synthesis and comparison. We then costed the 13 candidates the literature favours with Azure's Quantum Resource Estimator, including the cost most papers leave out: loading classical data into the machine.
None of the 13 clears the bar once that cost is counted. Quantum finance turns out to be an integration problem rather than a shopping list of algorithms. Microsoft supported the work with model and compute access.
- 2022 – 2023Academic · UPF BSc thesis · Belobaba VC
Could a shared ledger replace the way auditors confirm balances with third parties?
I became interested in blockchain when there were plenty of ideas about what it could transform and far less clarity about where it actually made sense. Conversations with people in the industry pointed me at accounting and audit, and at circularization in particular: the slow, manual way auditors confirm a company's balances with its customers, suppliers and banks.
With Gerard Merino Lago I explored it twice: first as a prospective venture at Belobaba VC, in a feasibility study presented to the investment committee, and then as our bachelor's thesis, extended into a venture plan with market sizing, operating model, go-to-market and financial scenarios. The lesson has stayed with me: start with the process and the problem, not with a technology looking for somewhere to be used.
About
I like figuring out why something works the way it does, and I am not very good at stopping at the first answer.
I started in operations, where I learned that the problem in front of you is rarely the whole problem. Wanting to fix those systems pulled me into product; wanting to build better ones pulled me closer to technology.
I have no loyalty to a technology. I care about whether it makes the system better. I have explored blockchain, gone deep on quantum computing, and right now I am focused on AI: agents, retrieval, context, knowledge systems and evaluation.
This is what I am working on, not what I am.
Work
- Feb 2026 – Jul 2026Microsoft (MDCC) · Dynamics 365 Business Central
I worked in product management for Dynamics 365 Business Central. A manufacturing video series became the starting point: I built reusable, AI-assisted tooling around how it was produced, so that a non-technical content specialist could carry it on without me.
From there I started AI work of my own on reliable agent use in Business Central, moving from how an agent interacts with the application, to the context and structured product knowledge it needs, to evaluating whole AI pipelines on quality, speed, usage and cost. It combined process design, prototyping and testing, always starting from where the system was failing and whether a different approach could improve it. It was research, not product: prototypes, evidence and a recommendation, and nothing shipped to production.
- May 2025 – Dec 2025Copenhagen School of Entrepreneurship · Event technology venture
This was where I became much more hands-on with software development. I took an operational problem I had lived through at European Blockchain Convention, validated it through 20+ interviews with event organisers and enterprise stakeholders, and turned it into a B2B SaaS MVP: an automated attendee-matching and meeting-scheduling system, with organiser controls for assignments, conflicts and last-minute changes, built through an AI-assisted development process.
It never launched and it is parked now. What it left me was the AI toolchain, and a clear sense of which of those tools hold up under a real product and which only look like they do.
- Jan 2023 – Feb 2024European Blockchain Convention
I worked on operations for a 4,000+ attendee conference across three venues, and became increasingly focused on improving the systems behind the event rather than just running them. The clearest example was sponsor matchmaking: a heavily manual process I redesigned with spreadsheets and email automation, coordinating 600+ meetings for 60+ sponsors with an 80% success rate.
I also managed 150+ media and ecosystem partnerships, increasing event coverage by 85%. The matchmaking system improved the process without fully solving it, and that gap became the starting point for the product I later built at CSE.
- Jun 2022 – Dec 2022Belobaba VC
Conversations with people in the industry led me to look at where blockchain actually made sense in accounting and audit. With a fellow student I co-developed a feasibility study for a platform to automate how auditors confirm balances with third parties, assessing the business process, the market alternatives and the technical architecture, and presented it to the investment committee.
We then extended it into a pre-launch venture concept and our bachelor's thesis. The same question, asked from the investor's side of the table.
Retail, alongside studies
- Jun 2017 – Jan 2026Burberry · Son of a Tailor
I started at Burberry alongside my bachelor's studies, in a customer-facing role, and progressed to Key Holder, supporting the store management team with day-to-day operations, reporting and team performance. I redesigned the reporting processes and built dashboards, and when COVID closed the store I led the move to remote selling, with a process the store kept afterwards.
At Son of a Tailor, alongside my master's, I designed the in-store customer journey, trained the team on it, and brought recurring customer feedback back to the product team. Selling something expensive to someone standing in front of you turns out to be customer discovery without the label. Those years are also where my interest in fashion comes from.
Studies
- Sep 2024 – Jun 2026Copenhagen Business School · GPA 10.9 / 12
I focused my electives on Applied Machine Learning and The Business Application of GenAI. My thesis asked whether quantum computing delivers a practical advantage in finance, and ran on an AI-supported thematic-analysis workflow that used several LLMs for coding, synthesis and comparative analysis.
- Sep 2019 – Jun 2023Pompeu Fabra University · TecnoCampus · GPA 8.2 / 10
A foundation in business administration, entrepreneurship and how innovation creates value, with an exchange semester at Management Center Innsbruck (MCI), Austria. My thesis explored where blockchain could meaningfully improve accounting and audit, rather than where it merely could be used.