Ottawa lab builds local AI platform on used servers

Christian Crete has built the NORA AI research platform in Ottawa using refurbished enterprise servers. The project focuses on locally operated artificial intelligence systems rather than externally hosted services.

Crete, Founder of NORA AI and CRETE Technology Group, is using the platform to test how local language models can work with physical computing infrastructure, organise technical information and support research tasks. He describes the system as an experimental environment combining local model inference, memory systems, tool integration and research automation.

Unlike many AI projects that rely on large cloud infrastructure or institutional backing, this lab has been assembled over time with second-hand and mixed-generation equipment, including HP ProLiant servers, Intel Xeon processors, NVIDIA graphics cards and Linux systems.

Crete said the project began with a basic question: could a single researcher build an AI system that did more than answer prompts? It has since developed into a broader effort to connect language models to surrounding systems so they can assist with observation, information retrieval and bounded technical workflows.

"I didn't start with a large budget, a university laboratory, or a team of engineers. I started with the hardware I could get, an idea I wanted to explore, and the willingness to keep learning. Every server has become another opportunity to test something," said Christian Crete, founder of NORA AI and CRETE Technology Group.

Beyond chatbots

A central aim of the project is to move beyond the standard chatbot model. NORA is being developed to examine whether AI systems can interact with equipment and software in a more structured way while remaining under human oversight.

A related agent environment called EVE is also part of the wider development effort. Together, the systems are being used to study how multiple AI components might support technical research, system observation, information management and controlled automation.

The platform remains experimental and does not treat automated workflows as equivalent to verified discoveries. Testing, logging and validation are part of the process, along with efforts to document limitations and check outputs.

"Chatbots are useful, but I'm interested in what comes next. Can an AI system understand the equipment around it? Can it help investigate technical problems, organize evidence, and suggest experiments? Those are the questions we're exploring with NORA," said Crete.

Local control

The emphasis on local infrastructure reflects a wider interest in direct control over computing systems and research assets. Running models locally can give researchers greater visibility into software configurations, hardware limits and operating conditions, according to the organisation.

At the same time, Crete said local deployment does not automatically resolve questions of privacy, security or reliability. Those issues still depend on system design, access controls, maintenance and operational practice.

That distinction matters in a sector where local AI is often presented as a simple answer to governance concerns. Here, the stated value lies less in claims of independence and more in the ability to inspect machines directly, observe constraints and test ideas without relying entirely on outside providers.

"I believe Canada needs people who understand how AI systems actually work, all the way down to the hardware. That includes small laboratories and independent builders, not just major companies. We need the ability to experiment, measure results, and develop our own technical knowledge," said Crete.

Broader research

NORA sits within a larger set of experimental programmes under CRETE Technology Group. These include XENOLINGUA, which focuses on software-defined radio, spectrum observation and signal analysis; VANTA, a broader collection of research initiatives; and SENTINEL, a defensive multi-sensor data-fusion project.

These efforts vary in technical scope and maturity, but they share a focus on practical system building and documented testing. In the radio programme, for example, the group uses software-defined receivers and analysis tools to investigate signals and create repeatable observation workflows.

The organisation does not claim independently verified scientific breakthroughs from these programmes. That caveat reflects the limits of operating with modest resources and without the scale typically available to universities, major laboratories or large technology companies.

Hardware remains one of the clearest constraints. The lab relies on older server processors and a mix of NVIDIA accelerators, affecting model size, inference speed, memory capacity and energy efficiency.

Crete said the project is seeking technical partnerships, equipment sponsorship and evaluation opportunities to expand what the lab can test. Areas of interest include hardware benchmarking, prototype work and controlled experiments using public or synthetic workloads, while keeping proprietary research assets on infrastructure controlled by the organisation.

"More powerful hardware would help, but the real goal isn't to collect expensive machines. It's to use the equipment responsibly, produce meaningful measurements, and learn something useful from every experiment," said Crete.

The immediate priorities for NORA include improving local inference, making research workflows more reliable, documenting system performance and widening technical collaboration. "I don't expect people to believe something just because I say we're building it. I'd rather show them the machines, explain the software, demonstrate what works, and be open about what still needs improvement. That's how you build credibility," said Crete.