SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
InnovaMat is seeking a Senior Computer Vision Engineer to own and advance the machine learning models powering Mathboard AI, a technology that reads handwritten mathematics on digital boards. The core challenge is building multiple-instance detection models that work directly on pen strokes rather than rendered images—a novel problem with no off-the-shelf solutions or public benchmarks.
You will take architectural responsibility for the handwriting-reading models, including input representation, instance proposal and assignment, training methodology, and continuous improvement. The work is genuine research engineering: diagnose where models fail, form hypotheses, implement solutions, and rigorously evaluate results. You'll design changes to handle hard cases, maintain challenger architectures to avoid local optima, and run controlled comparisons with pre-agreed success criteria.
A critical aspect is productionization: models must run fast enough on ordinary school hardware (not just GPUs) and behave identically in production as in evaluation. You'll monitor live performance on real student work and translate observations into the next iteration.
You'll collaborate closely with the dataset and evaluation owner to identify whether bottlenecks are model, data, or task-definition issues. The problem requires spatial reasoning and relational structure understanding—two strokes belong together based on proximity, scale, alignment, and context, not visual appearance alone.
InnovaMat has partnered with 27,000+ teachers across 2,800 schools in nine countries, reaching 600,000+ students since 2017. The role is based in Sant Cugat del Vallès, Barcelona, and offers the opportunity to solve a genuinely novel computer vision problem with real-world educational impact.
