Rob Imbeault on breaking the VC funding dry spell in Ottawa

February 2, 2026

Already an Insider? Log in

Get Instant Access to This Article

Become an Ottawa Business Journal Insider and get immediate access to all of our Insider-only content and much more.

There have been signs of resilience in Ottawa’s tech sector in the first weeks of 2026, including promising news from local startup Backboard.io. Rob Imbeault started off the year by announcing the closing of a substantial pre-seed funding round for his latest venture. This during what Imbeault, co-founder of Ottawa unicorn firm Assent, has described as a “dry winter” for venture capital. In fact, analysts have characterized 2025 as a perfect storm for Canadian VC, with only 21 funds closing across the country last year, raising a collective $2.1 billion, according to RBCx. The H1 2025 report from the Canadian Venture Capital and Private Equity Association showed $2.9 billion invested across 254 deals in the first half of last year, the lowest H1 total since 2020. It represented a 26 per cent decline in dollars invested and a 22 per cent drop in deal count compared to H1 2024, the report said. Average deal size held at $11.4 million, suggesting a sustained willingness to back strong companies despite fewer overall transactions, the report noted. Pre-seed and seed stages, however, continued to show a decline in activity. A combined $297 million was deployed across 133 deals, indicating compressed valuations and limited appetite for early-stage risk, the report concluded. According to Imbeault, the investment in Backboard — from Canadian venture firms Mistral Venture Partners, N49P, Garage Capital and Developer Capital — reflects his company's unique approach to a fundamental challenge in artificial intelligence. “AI today is powerful, but fundamentally stateless,” Imbeault told OBJ. “We’re building a permanent memory infrastructure that allows AI systems to learn and evolve with users over time.” Backboard functions as an independent infrastructure layer decoupled from the underlying large language model, or LLM. This allows a user to “hot swap” between more than 2,000 different models without losing context. For example, a user might start a task in ChatGPT-5.2 for ideation and finish it in Claude 4.5, with the second model retaining full access to the history established by the first. “This past couple of years, people are just kind of obsessed with this human brain metaphor for AI,” Imbeault said. “And I don’t feel it should work that way. I think it fails there, because human memory is not great, right? “A real memory layer turns AI from a stateless tool into a system that can build context over time,” Imbeault explained. “That is what enables things like long-running agents, enterprise assistants and products that actually improve the more they are used.” As a result, the platform that Imbeault is building as a solo founder at Backboard offers protection against the loss of institutional memory. “Say you’re using ChatGPT and your colleagues are all using ChatGPT for various functions,” Imbeault explained. “You have a CTO, you have a CMO, and when one of those individuals leaves, so too does all of that memory, right? Now with the Backboard platform, if the CMO leaves, you can use that memory to onboard the next CMO.” The validity of this approach is supported by the LoCoMo (long context memory) benchmark. Backboard.io recently secured top spot on the benchmark with a score of 90.1 per cent for accuracy. “The dataset is designed to examine not just an LLM but any LLM-based system’s capabilities and blindspots in a fine-grained manner,” explained Adyasha Maharana, the creator of LoCoMo and a research scientist at Databricks. “Raw human performance is somewhere around 88 per cent. Breaking the 90-per-cent threshold requires superhuman consistency in recall and reasoning.” The current generation of high-performing LLMs sits around 80 per cent in the LoCoMo benchmark, she noted. She explained how the "raw" performance of frontier models on LoCoMo is measured by feeding the entire conversation as a single prompt and asking questions based on that input. “Strictly speaking, this is not memory,” she said. “This is simply understanding whether the LLM pays attention to each part of its input and is able to reason over it correctly. The system built by Backboard.io is a far better attempt at simulating memory as it manifests in humans. It is practical, cheaper, scalable and doesn't rely solely on brute-force LLM processing for answers.” It’s perhaps no wonder, then, that Imbeault was able to attract an all-Canadian syndicate of investors as VC shifts focus from AI “wrappers” to foundational infrastructure. Garage Capital’s participation highlights how capital can be focused on hyper-specific elements of AI. The firm was an early investor in Groq, where the valuation recently skyrocketed to $6.9 billion. “I had no idea Garage was invested,” Imbeault noted. “They were the original investors in Groq. The windfall from Groq allows Garage to deploy capital into new infrastructure plays.” Imbeault’s return follows his exit from Assent, which he helped scale from a basement startup into a global supply chain platform valued at $1.4 billion. He spent a short time as an angel investor but felt compelled to return to building. “You can only sit on so many beaches,” Imbeault said. “Why I’m a founder again is more personal, but let’s say I want to be a part of this AI wave.” His approach to fundraising reflects his experience as a founder. “I did not present one deck. I did not talk to one slide, this whole thing,” he said. “Instead I relied on a technical memo and direct conversation.” The exact value of the raise remains undisclosed, but Imbeault confirmed it is “in the millions, but less than $10 million.” He was candid about the company’s current stage, stating it has not yet found a product-market fit. “I guess it’s more of a technical SaaS thing,” he said. “I want traction. That’s why we raised the money. The primary goal of 2026 is to get to the point where we’ve figured out the formula for scale. That’s what we’re truly aiming for.”