From Classroom to Production: Building Edge AI Infrastructure with EdgeLake and UC Santa Cruz
Author: Roy Shadmon, PhD / System Architect, AnyLog
Over the past two years, we at AnyLog have worked with UC Santa Cruz to turn student capstone projects into a collaboration that is directly influencing EdgeLake capabilities, open source projects, and real-world deployments.
For me, the collaboration also has personal roots. It started during my PhD at UC Santa Cruz, where I served as a teaching assistant for the two-quarter software engineering capstone series, CSE 115B/C, taught by Dr. Richard Jullig. In the course, teams of three to five upper-division students spend six months working with industry sponsors to turn ideas into real-world value.
Giving Students a Faster Path from Idea to Application
After supporting the course several times, I saw that one of the biggest constraints students faced was simply getting development environments working consistently across every team member’s machine.
Students were developing on Macs, Windows PCs, and Linux systems while configuring databases, aligning schemas, ports, networking, and data sources, and troubleshooting environment-specific issues. In a six-month capstone, that friction could consume a disproportionate amount of time and would limit how far the actual idea progressed.
EdgeLake changed the starting point. With a one-click deployment, each student could create the same digital twin of the target environment locally, with the data layer, distributed services, and a single REST API ready for the application to interface with the distributed environment. Teams could also launch additional EdgeLake nodes on the same laptop to simulate separate machines, sites, and data sources or utilize each other’s laptops as a distributed cluster environment.
That gave everyone a consistent environment from day one and allowed students to move much more quickly into building, testing, and extending their applications.
Just as importantly, that consistency carries directly into production. The same EdgeLake deployment model used on student laptops can move onto industrial PCs, edge devices, and servers, with the application simply pointed to the production endpoints. The result is faster development, stronger proofs of concept, and a direct path from local simulation to real deployment.
From Student Projects to Real-World Edge Innovation
Over the past two years, AnyLog has sponsored five student projects spanning federated learning, decentralized video streaming with embedded AI inference and searchable metadata, the AnyLog Edge Data Manager, local microgrid management, and distributed power-utility infrastructure.
The collaboration has evolved from using EdgeLake as the foundation for student PoCs into a way to develop capabilities that AnyLog continues to extend, productize, and deploy in real environments, all without requiring a centralized cloud dependency. Several projects illustrate that progression:
Smart Home Energy Management
Students are using EdgeLake as the local data and intelligence layer for a home microgrid, unifying data from solar generation, battery storage, energy consumption, and connected devices without relying on the cloud.
EdgeLake gives the application a single API to these distributed resources so it can monitor conditions and support intelligent decisions and recommendations about how energy is generated, stored, and consumed. The project now runs on a professor’s home in the Santa Cruz Mountains.
Real-Time Grid Intelligence and Utility Outage Detection
Another project extends the concept from a single home to distributed utility infrastructure.
The project explores how utilities can monitor distributed grid infrastructure and identify the source and impact of outages in real time. EdgeLake keeps data and processing distributed across utility assets while presenting them as a Single System Image, allowing the application to query and correlate information across the grid without first centralizing the data.
The work has progressed into an advanced PoC and is being developed into a demo and testbed for utilities to evaluate this distributed architecture.
Edge Data Manager (EDM)
Edge Data Manager is the graphical interface for interacting with an EdgeLake deployment as one cloud-like system, regardless of how many machines, sites, databases, or data sources sit underneath it.
What began as a student project has evolved into AnyLog’s primary management interface for querying data, monitoring nodes, exploring distributed resources, and managing EdgeLake deployments.
EdgeFL and Federated Learning
EdgeFL uses EdgeLake to coordinate federated learning across distributed sites without moving raw training data to a central location.
Each site trains locally, EdgeLake provides the distributed data and coordination layer, and the resulting models are shared across the network and deployed for inference at the edge. This creates a continuous path from managing distributed data to training AI models to executing inference, all at the edge.
The project has since evolved into the open source EdgeFL framework, and IBM Open Horizon built an automated deployment around it.
Giving Students Real-World Open Source Experience
One of the most valuable outcomes of this collaboration is the experience students gain building software that is expected to work beyond a classroom setting.
Instead of stopping once a feature works, students have to think about code efficiency, failure modes, reliability, scalability, deployment, and what happens when a system operates continuously in a real environment.
They also learn that building a useful product is iterative: understand the problem, engineer a solution, test assumptions, identify what breaks, get feedback from end users, and improve the system until it delivers real utility.
The strongest outcome is that their work does not necessarily end when the course does.
The Edge Data Manager evolved into the primary graphical interface used with EdgeLake deployments. Federated learning work became EdgeFL. Work on decentralized video streaming, inference capture, and search was incorporated into EdgeLake itself. Other projects, including microgrid management and utility outage detection, have progressed into real deployments or advanced PoCs.
Student work has directly influenced EdgeLake and the applications being built on top of it.
What’s Next
The next phase of our collaboration with UC Santa Cruz is focused on expanding the strongest projects into broader research and real-world deployments.
The smart home microgrid pilot aims to grow from a single home to multiple homes in the Santa Cruz Mountains, and we welcome others interested in participating to reach out. The utility project is being developed into a demo and testbed for utilities to evaluate a distributed architecture. EdgeFL continues to mature, with the goal of bringing the architecture into UC Santa Cruz research AI labs.
AnyLog also plans to continue sponsoring new student teams whose work can contribute directly to EdgeLake and its growing set of applications.
“Our work with students and academia is ultimately about expanding what the next generation of engineers believe is possible with decentralized architectures and giving those engineers the tools to build applications where data, compute, and AI operate together at the edge,” said Moshe Shadmon, Founder and CEO at AnyLog
Industry has spent decades designing around centralized data centers, but the rise of AI and physical AI is changing that model. Real-time intelligent systems cannot depend on a single location or network connection. Intelligence needs to operate where data is generated and decisions are made, without needing to move data away from the operating environment to utilize it.
We want students not just to learn distributed systems, but to help shape the next generation of resilient, decentralized applications with no single point of failure and true real-time intelligence.
This is just the beginning of the story. In the coming weeks, we’ll take a closer look at each of these student projects, exploring what the teams built, how they’re using EdgeLake, and what they learned along the way.
