STEMilie

Inspiration

Our project stems from a personal need: as international students, we have experienced firsthand how difficult it can be to navigate bureaucracy, regulations, and logistical procedures in a new country. Often, it’s unclear whom to turn to for practical questions, and the risk is getting lost among dozens of websites, PDFs, and regulations.

What it does

That’s why we envisioned a digital assistant acting as a “civic friend”: an accessible, transparent, and reliable chatbot that provides clear answers to questions about permits, registrations, tuition fees, transportation, and more. Our vision is to connect the assistant to a phone number / WhatsApp, so you can interact with it just like a real friend ready to help you.

How we built it

Using RAG assistant structure, HTML and PDF files with preprocessing, chunking, embedding, and unifying backend and frontend using Streamlit and the transparency of Apertus open source.

Challenges we ran into

Implementing audio, conflict for different implementations

Accomplishments that we're proud of

We are proud of having developed a RAG-powered assistant that is able to provide information about bureaucracy and other matters as well that concern three cantons of Switzerland. It is cost-optimised thanks to the of LocalStack, a software that simulates cloud services locally, allowing us to test and validate without having cloud service costs. We also implemented Airflow to orchestrate the update pipeline, using a DAG that systematically checks for updates and data integrity. It easily scalable up to all the cantons as we would just need to upload the officials documents from which out chatbot would take information.

What we learned

To build a chatbot, to handle different challenges like audio, adding a front-end using Streamlit

What's next for STEMilie

We hope to cover all the cantons as well

Built With

• jupyter
• python