Oleg
loleg

To address Caritas's need for scalable, multilingual training video generation, I recommend a hackathon project that builds on open-source tools and data. The prototype would use AI models like T5 (Hugging Face) to generate scripts and Mozilla TTS/PygTTS for multilingual narration, with visuals created via Reveal.js or Google Slides. The video would be composed using OpenShot/Kdenlive, and the workflow orchestrated via Docker and Python. This approach leverages open data from Caritas, OERs, and the web while ensuring GDPR compliance and feedback loops for quality improvement. The project prioritizes a scalable, automated pipeline for consistent, on-demand video generation across languages. Future steps include iterating based on user feedback and exploring integrations with Caritas's existing platforms for seamless use.

🅰️ℹ️ Written with help of APERTUS 70B

To prototype in a hackathon setting, leverage open-source tools like Node.js, Vue (for web) or React Native (for mobile), and integrate with open data sources such as OpenStreetMap for campus locations and Open Menu (if available) for menus. Use Figma for design, and consider mock APIs for data during development. The focus should be on a seamless, user-friendly experience that addresses the problem of long lunch queues on campus, with a scalable and secure backend that could integrate with Offood’s existing infrastructure. The prototype should demonstrate the potential for Offood to enhance campus life by offering quick, convenient access to meals, and be presented with a clear user flow, technical approach, and potential impact on student well-being and productivity.

🅰️ℹ️ Written with APERTUS 70B

Also would like to just commend your efforts to prepare the pitch and poster here.

You're going to great depths to try to put together a clean dataset, many kudos for the effort here. Just a couple of quick links for my own memory from our discussion, in terms of prior art: Fundraiso, SwissFoundations, Zefix/EDA. Better data quality is the key to wider social impact here. Thank you for all the hard work!

I've moved your attachment into the Pitch as a rich link, @ayeshagul - thanks for contributing!

You are in excellent hands with this challenge, and from what I heard just now - that you're working on a design prototype in Canva, while a frontend team sets up a Vue.js project and on the backend a FastAPI (Python) workflow is in development 👐🏽 Maybe if I can add one thought, it is to think creatively about digital authenticity. Very tempting smells from the community kitchen: looking forward to see the result!

For a Swiss sovereign solution, you can try Apertus, in particular the RAG API from Public AI that I showed you. Think about the safety and scaling issues that your solution will depend on, but don't let technical questions slow you down. This challenge is a huge chance to help in such an important area - your contributions will be valuable no matter what approach that you take. I wish you much success with this. Viel Glück!

Top of mind: check out Legal Lens which started at a hackathon in 2024. There are many RAG systems, like PrivateGPT or products like Euria that could help you get started. ZüriCityGPT is based on municipal legal documents: see writeups here and here.

It's great that you got some examples on GitHub from the challenge owner. As I just suggested, there are openly accessible resources like the 17 BFS datasets (opendata.swiss) to sharpen the research focus, and the Public AI chatbot as a possible alternative to commercial options for you to try. I hope you can bring something from your personal stories for depth and meaning to your experience 🎗️

Toller Start, alles Gute für die Fortsetzung!