Technical Product Manager

• Own the roadmap for a model-powered product surface, from customer problem to shipped feature

• Read eval reports and confusion matrices well enough to argue with the CV engineers, not just relay their conclusions

• Make and defend accuracy/speed/cost tradeoffs — in the room with engineering, and directly with customers

• Build and maintain eval sets that keep the team honest about model performance

• Decide when a model is actually ready to ship — not based on aggregate accuracy, but on which failure modes are cheap to catch downstream and which are expensive to miss

• Turn ambiguous customer complaints ("it missed something") into a specific, testable model or UX fix

• Work daily with our CV and ML engineers as a peer who understands the model, not a translator layer

• 4+ years in product, with real ownership of a model-powered feature (not just "worked with" a data science team)

• Fluent in model evaluation — precision, recall, false positive/negative tradeoffs — and can reason about when "85% accurate" is good enough versus when it isn't, based on the cost of the specific errors, not just the aggregate number

• Has shipped 0-to-1 or a major iteration on an ML/CV-driven product in B2B SaaS

• Comfortable being the last line of defense on "will this model actually work" for customers and execs

• Fast learner — no construction background required, but you'll need to be dangerous in it within 30 days

• Ships fast, iterates in public, doesn't hide behind process

• Prior experience specifically in computer vision (vs. NLP/LLM-only)

• Background at a company like Scale AI, Labelbox, Samsara, Matterport, Cape Analytics, or DroneDeploy

• Comfortable in Figma and reading code, even if you don't ship it

We are proud to offer competitive, top-of-market compensation because we want to celebrate the dedicated people who ship amazing work and drive our success. Our individual compensation is thoughtfully tailored based on your role, experience, and contributions, alongside performance-based rewards.