Sr. AI Researcher, Agentic Systems

At OpenSpace, we’re redefining how the world’s most complex projects are built. Our AI-powered Visual Intelligence Platform uses computer vision and spatial AI to give construction teams a real-time view of what’s happening on-site, helping them build faster, safer, and with greater confidence.

But what truly sets us apart is our people. We hire curious, driven teammates who love solving hard problems, taking ownership, and making a real-world impact. Great people build great culture—and apparently it shows. Forbes has named OpenSpace one of America’s Best Startup Employers five years in a row. Come see what all the fuss is about ✨

The role:

The Sr. AI Research Engineer will build agentic systems that reason across the full OpenSpace data record: 360° imagery, point clouds, BIM models, drawing sets, and specifications, plus the newer streams coming off mobile capture, wearables, and field audio. Our computer vision team builds the perception primitives that determine what is in a capture. You will build the layer above, combining those primitives with frontier models to answer the questions project teams actually ask, and making the answers verifiable enough that a builder will act on them.

The work is applied research. The hard problems are deciding what an agent should see out of a four-dimensional site record, designing the tools it can call, and proving it is right often enough to ship. By the time you are calling a frontier API, the interesting decisions have already been made.

What you will be doing:

• Design and build agent systems that plan and execute multi-step tasks across imagery, geometry, and documents.
• Own the context layer. Build retrieval and grounding over a spatially and temporally indexed site record so an agent gets the evidence a question actually needs.
• Design the tool surface: the primitives an agent calls to locate a component, compare two captures, read a region of a drawing, or query a BIM element.
• Build evaluations for tasks where ground truth is ambiguous, then use them to make honest calls about what is ready and what is not.
• Choose the right model for each step. Frontier API, open-weight model, fine-tuned small model, or classical CV, weighed against cost, latency, and accuracy at production volume.
• Make outputs traceable. Every claim should point back to the pixel, sheet, or spec section that supports it.
• Prototype quickly, kill what doesn't work, and partner with platform engineers to ship what does.

What we are looking for:

• M.S./Ph.D. in a related field or equivalent industry experience.
• 5+ years building ML systems, including at least one LLM or VLM-based system you took to production and kept running. We will ask what broke.
• Experience designing evaluations from scratch where ground truth is expensive, sparse, or contested.
• Strong multimodal fluency: vision-language models, embeddings and retrieval, context design, and fine-tuning when it earns its cost.
• Solid engineering. Python, async orchestration, tracing and cost instrumentation. You build your own harness.
• Comfort with spatial data. You don't need SLAM depth, but coordinate frames, camera poses, and projections shouldn't slow you down.
• Judgment about what is worth automating, and the willingness to say when the answer is nothing yet.

Nice to have:

• Post-training or reinforcement learning on agent trajectories.
• Document understanding over technical drawings, specifications, or other structured PDFs.
• Sensor streams beyond core imagery, such as mobile capture, wearables, or field audio.
• Experience in construction, robotics, geospatial, or another domain where models meet the physical world.

Base Salary: $181,000 - $230,000 CAD