real-world projects
A growing collection of real-world deployments and experimental interfaces.
vTaiwan Agentic Citizen Assembly
I teamed up with the vTaiwan community in Sep-Oct 2025 for an experimental project. It’s an exploratory inquiry to see what the future might look like, rather than a formal tool for making high-stakes decisions right now.
In Taiwan, democratic deliberation often struggles with barriers like time, geography, and resources, meaning a lot of it happens only in major cities. This leaves many voices out of the loop due to structural constraints.
So I asked: Can AI agents, modeled after real participants’ values and stances, help bridge this gap? I built a system where users can spin up “Sim-like” agents to debate and collaborate in virtual spaces, simulating the presence of people who usually can’t make it to the table.
To be clear, I’m not suggesting that we replace human agency with automation. Instead, it’s a critical look at what this kind of future would entail. Democracy is a practice, not just an outcome!
Interactive Demo
Kultur Komitee Winterthur
Kultur Komitee is an ongoing, real-world budget assembly in Winterthur. Every year, residents come together to deliberate and collectively allocate 400,000 CHF to fund local art and cultural projects — and I use each cycle as an opportunity to experiment with different voting frameworks and allocation rules!
Most of my work focuses on iterating over the Method of Equal Shares (MES). I design and test new interfaces and algorithms to make the voting process fairer and more transparent. Together with my colleague Fynn Bachmann, I have been running the calculations and aggregations for the 2024, 2025, and 2026 editions — collectively shaping how over 1.2 million CHF have been allocated to art and cultural projects in the city.
For instance, in the 2025 deployment (KK25), I worked with 38 residents to evaluate 121 proposals. They deliberated in groups and cast individual point votes, which my platform merged into a single fair allocation complete with personalized vote receipts.
In 2026 (KK26), I leveled up the framework even more. I added new features that let voters see exactly how different aggregation rules would change the final results. The platform breaks down how each individual vote translates into actual spending, giving everyone unprecedented transparency into the budgeting process.
Interactive Dashboard (KK26)
Stadtidee Aarau Participatory Budgeting
In 2023, the city of Aarau launched its very first participatory budgeting program, Stadtidee Aarau, distributing a 50,000 CHF budget among 33 proposed projects.
The real-world setup was closely informed by preliminary results from our lab experiments. Working in close collaboration with Sebrina, Jasmin, and Lea from the city of Aarau, we planned every step of the implementation together — from how people would vote, to how the results would be communicated back to residents. The city decided to use a 10-point voting system and aggregate votes using the Method of Equal Shares (MES).
About 1,800 of Aarau’s 22,000 residents ended up casting their votes. To help voters understand what MES actually does, the city website explained how the algorithm works before voting opened. Afterward, it showed a breakdown of how the budget was divided across districts. The result: MES funded 17 projects spread across all districts, compared to just 7 (on average more expensive) projects that a standard “most votes wins” approach would have chosen.