AI Product Manager, Safi

• You must be comfortable planning with ML engineers, reading model-performance data, and reasoning about trade-offs with people
• This is a junior-mid level role for someone who is excited about deploying AI that impacts real-life industrial plants
• ML literacy: you have a basic understanding of how computer vision AI works
• Data analysis: you have experience analysing data and taking accurate conclusions
• Great people skills and confidence: You will need to walk onto an industrial site and be able to connect with forklift operators and plant managers, some of whom will not be able to speak the same language as you
• Information absorption: you will need to be able to quickly absorb information about materials, business needs, sales requirements and make a plan
• Able to write and maintain detailed documentation for your team. This is the core of taxonomy management
• Genuine curiosity about the physical material and how it's handled
• Resilience: willing to travel extensively to recycling and processing sites in un-glamorous parts of the world
• If our mission resonates with you, we encourage you to apply, even if your experience doesn't match every requirement

Desirable

• As an AI Product Manager for Vision Models, you are the critical layer that connects on-the-ground AI performance with our AI/ML engineering
• Capture and triage data + feedback from customer deployments
• Use your judgement and understanding of our customer's industrial processes + priorities, recycled material, and our vision AI capabilities to prepare and prioritise work for our AI/ML engineers
• Lead the domain discovery required to bring new materials and markets online
• Own the data pipeline and labelling teams that we use to teach our models
• You will spend time at recycling sites watching how operators actually assess quality - and translate what you see into precise AI model development, labelling and training sets
• We think much of this role can be learnt on the job - high energy, commitment, quick thinking and willingness to get your hands dirty (quite literally, we work with waste materials) will go far
• Own the ML backlog and prioritisation. Triage incoming data and feedback from customers and colleagues. Assess with sales, deployment and customers to understand importance. Execute the work - training preparation, data collection, labelling - that our ML engineers need to solve the customer problems
• Review model feedback and spot trends. Regularly review incoming AI-model feedback, identify recurring issues or trends, and feed the significant ones into the ML triage process
• Maintain and curate the labelling guidelines (taxonomy). Keep the guidelines accurate and up to date as understanding improves and new edge cases emerge. Decide when to split out a new class and when to fold classes together — decisions that directly determine final model quality
• Lead new-material and new-partner model development. Understand deeply with our customers how different objects, contamination and material affects their process. Understand what matters when assessing its quality: what can be judged visually, how accurate detection needs to be, and where the hard cases are
• Gather detailed specifications, convey them to the ML team, sanity-check feasibility, create the labelling guidelines, and get annotators onto the work. Run weekly reviews with partners
• Act as the interface between the business and the ML team. Run the weekly cadence with ML engineers, deployment, and sales so that priorities, guidelines, and progress stay aligned
• We're a small team and we all automate our own work - most people here use LLM and agentic tools to streamline triage, reporting, and recurring processes. We use Linear, Notion, Claude, GitHub and GCP. If you have a builder's instinct for making your own job easier, you'll fit in well