At the Port of Montreal, where containers move on schedules dictated by tides, rail slots, and trucking windows, the math of getting a single box from ship to gate is brutal. It is the kind of problem that has historically belonged to operations researchers, the people who built airline crew scheduling and warehouse routing long before the current wave of machine learning. Funartech, a Montreal startup founded in 2017, is betting that the most useful industrial AI for customers like the Montreal Port Authority is not pure ML at all, but a hybrid that stitches machine learning together with the older discipline of operations research [Funartech].
That thesis, which the company calls the hybridization of ML and OR, is the entire wedge. Funartech describes itself as a Montreal-based startup specialized in providing services in AI, and notes it is also working on its first products [Funartech]. For now, the business is project work: the team listens to an industrial customer, reviews the relevant academic literature, and then implements customized algorithms rather than dropping in an off-the-shelf model [Funartech]. One of the company's stated mottos is, plainly, "where machine learning meets operations research" [Funartech].
The pitch is less exotic than it sounds. Operations research is what you reach for when a problem has hard constraints, a truck cannot be in two yards at once, a furnace cannot exceed a temperature, a crew cannot work past a legal limit. Machine learning is what you reach for when a problem has messy data and patterns no human has bothered to write down. Most real industrial problems have both. Funartech's argument, articulated by co-founder and CEO Dr. Nikolaj Van Omme, is that combining the two methods goes further than either alone [Funartech]. Van Omme has discussed the approach publicly, including a podcast appearance framed around using artificial intelligence to address climate-related industrial problems [Alain Guillot Podcast].
The bet
Funartech's commercial wedge today is bespoke industrial AI services for customers who have a specific optimization problem and the patience for a research-grade engagement. The Montreal Port Authority is the most visible reference. In a testimonial posted by the company, the MPA called Funartech "one of the best startups" it has worked with, describing the team as "a very mature and attentive startup that continually seeks to surpass itself" [Funartech testimonials]. Trade press has also reported that the company convinced Japanese automotive supplier Aisin of the merits of its hybrid AI approach [Inyulface], a notable validation for a small Quebec shop trying to sell into global manufacturing.
The company is publicly associated with AI Launch Lab, an incubator-accelerator in the Montreal ecosystem [Funartech partners], and has appeared as a partner on the Viva Technology platform [Viva Technology]. It is not, on the public record, a venture-funded scale-up chasing a Series A narrative. It looks more like a research-led services firm trying to convert deep technical work into productized IP over time, the way a number of European industrial AI shops have done.
Why it could be big
There is a real market gap underneath this. Most enterprise AI vendors of the last three years have leaned heavily on large language models and computer vision, which are powerful for text, image, and unstructured data tasks but not naturally suited to constraint-heavy scheduling, routing, and resource-allocation problems. Those problems still run, in many large industrial buyers, on aging OR codebases or on spreadsheets. A vendor that can credibly bridge ML and OR for a port, a manufacturer, or a logistics operator is selling into a category that the current AI hype cycle has largely walked past. Funartech's framing of climate and societal impact projects [Campus Party] also aligns with how European and Canadian industrial buyers increasingly scope procurement.
The team
Funartech was co-founded by Dr. Nikolaj Van Omme and Dr. Dania El-Khechen [Inyulface]. Van Omme, the CEO [Alain Guillot Podcast], holds an MSc in pure mathematics, an MSc in theoretical computer science, and a PhD in applied mathematics and industrial engineering [MTL connect], a stack that maps almost exactly onto the ML-plus-OR thesis the company is selling. El-Khechen holds a PhD from Concordia University, completed in 2009, on decomposing and packing polygons, a classic computational geometry problem with direct industrial applications [Concordia University]. Her published research record includes eight scientific papers and two highly influential citations on Semantic Scholar [Semantic Scholar]. For a deeptech services firm whose differentiation is the quality of the algorithmic work, founder pedigree of this kind is the asset.
The honest counterfactual
The bear case is straightforward. A research-led services business is hard to scale: each engagement is bespoke, margins compress as senior people get pulled into delivery, and the leap from services to a repeatable product is the move that kills most consultancies that try it. Funartech itself flags that it is still working on its first products [Funartech], and the company has not yet published the fundamental research it describes on its R&D page [Funartech]. The bull answer, supported by the cited evidence, is that the customer roster Funartech has assembled at this stage, the Port of Montreal as a named reference [Funartech testimonials] and reported work with Aisin [Inyulface], is exactly the kind of anchor-customer base from which a productized offering can credibly be carved out. Industrial AI products tend to emerge from a handful of deep deployments, not from a cold-start product launch.
What to watch
The most important milestone over the next twelve months is whether Funartech ships the first of the products it has been signaling on its website [Funartech]. A second signal worth tracking is publication: the company has said it intends to publish its hybridization research [Funartech], and a peer-reviewed paper would do more for enterprise credibility than any marketing site could. A third is whether any of the project engagements convert into multi-year platform contracts, the structural shift that would tell the market this is a product company in the making rather than a boutique.
For now, the patient population, to borrow a phrase from a different beat, is industrial operators with hard constraints and messy data, and the standard of care is still a patchwork of legacy optimization software bolted to newer ML pilots. Funartech is making a specific, technically grounded argument that those two worlds should be one. The next year will show whether the argument scales beyond the harbor.
Pulse Raman, Startuply