Profiles in IT: AI Makerspace with Joe Finnegan
The College of Science and Engineering's AI Makerspace is a hybrid physical and digital learning space where all University of Minnesota students, faculty, and staff can learn, engage, and create with AI. We interviewed the manager of the AI Makerspace, Joe Finnegan, about the impressive student projects and AI resources available for staff through the makerspace.
How can IT professionals use the Makerspace as a development or test environment for their innovations?
This is an area we can really shine for staff users. We are designed to be a safe sandbox environment for all kinds of development and testing - not just student or internal projects. We can provide shared access to great systems for local end-to-end testing and development of AI tools - we currently have four DGX Spark systems which are each capable of running Large Language Models (LLM) locally with tens to hundreds of billions of parameters locally. We also have a couple general purpose Desktops with NVIDIA Graphics Processing Unit (GPU) that are great for development and testing. Our systems are completely isolated from larger University systems such as data storage or accounts, so it’s a much safer place to try new things and even make mistakes as you go. When you have a good working proof of concept, we can then support you in securing and formalizing your tool and help you get it ready for review by groups such as University Information Security (UIS) if necessary. Dive into the sandbox: get your hands dirty, break things, fix them, learn, play, and explore safely.
What specific guardrails or guidelines does the AI Makerspace enforce to ensure sensitive University or student data doesn't accidentally train public commercial models?
We enforce (and educate people on) the University’s existing policies around the use of Generative AI tools and data safety - this means that Enterprise Gemini (logged in with your @umn.edu) is the only acceptable Generative AI tool for use with University Data. Our contract with Google stipulates that they cannot use our data for training purposes, but we do not have that same arrangement with any other AI service providers to my knowledge at this time. So - it’s pretty simple: if it’s University data - especially if it’s something like FERPA, you can only use Gemini with Enterprise-Grade Data Protection. Microsoft Copilot is also available for University use but is not approved for use with Private-Highly Restricted data like FERPA data. Also, it’s important to note that using another service’s “don’t train on my data” setting does not mean that sensitive University data can be used in that tool - it’s not the same thing as the robust legal agreement we have with Google that explicitly addresses data integrity and security.
What are some of the most technically impressive or innovative projects you’ve seen coming out of space so far?
One of the most technically impressive student projects I have seen in the makerspace so far is an intelligent drone swarm control mechanism. To even start a project like this takes a really high-level understanding of control systems, path planning, radio communications, single packet-level networking, Printed Circuit Board (PCB) and hardware fabrication, and domain-specific programming languages. What this student is essentially creating is a massive, decentralized multi-agent control system that can scale to many use cases. A framework like this could eventually be adapted for use in all kinds of industries; controlling a group of drones is just the first practical, real-world manifestation of the underlying system they are building. It is unbelievably cool to watch this 100% student-conceived and driven project come together.
Another really impressive student project is a tool called UConserve that scans any Universal Product Code (UPC) barcode and returns an extremely robust and detailed environmental impact summary for that product. And I really mean detailed - down to things like the chemicals used to pave the roads that the trucks moving this product or its derivatives drive on. This project combines multiple different large-source databases and performs an (Environmental Protection Agency) EPA Tool for Reduction and Assessment of Chemicals and Other Environmental Impact (TRACI) scan of the complete supply chain for the input product(s). The tool then uses the data sources and evaluation tool to calculate the CO2 equivalent per US dollar (kg CO2 eq / USD) of that product. This measures the amount of greenhouse gas emissions associated with each dollar of spending on a product or sector. The fact that a student built a supply chain analyzer at this level of granularity in the first few months of the AI Makerspace being open is a massive achievement for them and something we’re really excited about.
How can a team or individual get started scheduling a consultation or submitting a project proposal to work with the AI Makerspace?
Anyone is welcome to walk-in or schedule a brief consultation with our staff where they can offer their expertise to help you plan out or conceptualize your ideas into projects. The AI Makerspace does have development capacity and there are a couple ways to engage. Right now, anything requiring staff time, can be submitted via the AI Coordination Committee Consultation request request form. If approved, the AI Makerspace can partner with CSE-IT to develop and support projects. We are also currently in the process of opening an ISO where departments will be able to purchase AI development and consultancy services directly from the AI Makerspace and we’ll post more information about that on the CSE Makerspace website when things are in place.
Partner with the AI Makerspace
If you would like to partner with the CSE AI Makerspace to hold workshops, guest lectures, or provide course content. Please email [email protected]