A few weeks back me and @meritonaliu walked into Hack Winterthur 2026 as two CS students from ZHAW.
We walked out the winners, being able to provide actual results: Winterthurs LocalMate: A digital assistant that builds the bridge between information and the people that need it.
In this blog I want to outline the story of how it happened, why it matters to us personally, and whats under the hood.
The Challenge
The City of Winterthur proposed a straight forward problem: They want to strengthen the communication between their local government and the people they govern.
A simple problem in theory, but in practice something that is hard to implement: People get lost in the infinite amount of documents, regulations and the complex German.
This problem isn't abstract to us. We know what it's like.
Why this is personal
We both grew up in Zürich as the children of migrants. We watched everyone around us struggle with letters they couldn't fully understand, systems that assume familiarity with the system and other barriers that make integrating into this society harder than it should be.
We often built this bridge: Translating, explaining and navigating where we could. That experience shaped how we think about the accessibility of information.
And seriously: It shouldn't require a native speaker, a law degree or a teenager at your diner table to figure out how things work.
When we saw the challenge, we didn't need to brainstorm: Winterthur faces a challenge we faced our entire lives.
48 Hours of Building
Hack Winterthur wasn't just about hacking: We started with a simple prototype to start gathering insights as fast as we could.
The schedule was intense:
- Building a prototype during a 2 hour train ride home from 11PM to 1AM.
- Deploying everything on a personal Proxmox server.
- Getting user feedback as quickly as possible by using every time between document indexing as an opportunity to walk around and let people try it.
The atmosphere was electric. Entrepreneus from all sorts of fields (IT, Business, Fintech, etc.) all building under the same roof.
The perfect opportunity for cross-pollination of ideas.
What LocalMate actually is
LocalMate is a conversational AI assistant that answers questions about the City of Winterthur, accurately, simply and in any language.
Additionally it's an agent that retrieves dynamic information on the fly: Ask him about the trash schedule and he goes retrieving on his own.
You can try it right now: LocalMate MVP
The Architecture
Under the hood, LocalMate is built on a RAG Pipeline (Retrieval Augmented generation) with agentic workflows built in.
Data Ingestion
We scrape data from stadt.winterthur.ch, the official website of the City of Winterthur. This ensures that no information is wrong or hallucinated.
Vectorization
Everything is vectorized in memory, enabling extremely fast semantic search.
Retrieval
Before every request the users question was enriched using a small model and translated into german to guarantee retrieval quality.
For retrieval we used SBERT, a semantics aware transformer that allows for incredibly quick lookups that know what the user wants.
Multilingual by default
The system can retrieve and respond in whatever language the user wants: Ask in Albanian, get an answer in Albanian.
What makes it different
While most websites have an FAQ section or a simple chatbot, we offer something more.
- Verified Sources only: Every answer traces back to a document or page the city provides. No hallucinations.
- Language-agnostic: It doesn't just translate, it understands the semantics and reformulates in the users language
- Plain language: Administrative jargon gets simplified without losing accuracy.
- Agentic: Beyong simple Q&A, LocalMate can autonomously fetch dynamic data like schedules.
What's next
Winning Hack Winterthur was the starting line, not the finish.
WinLab
We have submitted LocalMate to the WinLab co-creating platform, seeking collaboration with the city to pilot the tool on a larger scale.
Launch Control Incubator
We're participating in the Home of Innoviation's incubator program to develop our understanding for business and creating.
Beyond Winterthur
The architecture is instance-agnostic. Any municipality, canton, public- or private-sector organization that has tons of public or private knowledge could get their own LocalMate instance.
We're actively looking for early adpoters.
The takeaway
The best hackathon projects solve a problem the builders actually have. For us, LocalMate is something we wished existed growing up.
We're building it so the next family doesn't have to figure it out alone
Me (Anes Hodza) and Meriton Aliu are students at ZHAW.
We publish our side projects under the name Centro Labs.
Originally featured on the ZHAW News Blog.
Top comments (0)