Head of Machine Learning and Data Science
Posted Updated
Description
- The Machine Learning and Data Science team builds and maintains the computer vision models at the heart of Winnow’s product: systems that watch a kitchen bin and turn what they see into accurate, trustworthy waste data
- We’re looking for a Head of Machine Learning and Data Science to take full ownership of one of the most technically challenging areas in the company
- You’ll lead the team, set the technical direction, and make sure our models keep getting better as Winnow scales into new sites, new cameras, and new food environments
- Your first priority is clear: take food recognition to the accuracy our customers need, which means bringing rigour to data quality, labelling, evaluation and ML operations
- From there, your remit extends to production forecasting, and over time to shaping how we use AI more broadly across the product, including LLM-based features
- This is a hands-on leadership role for someone who wants deep technical ownership and direct impact, rather than a large organisation
- Reporting to the Head of Technology, you’ll be a key part of the technical leadership of Winnow
- You’ll make the case for where we invest in ML, and be the bridge between ML and the rest of the business - translating model performance into terms that product, engineering and customer-facing teams can act on
- As the function matures, we expect this role’s remit to grow beyond the initial scope focused on food recognition
- Strategic ownership - develop and execute a long-term vision of creating a state-of-the-art AI/ML ecosystem. Influence how AI and ML are used in Winnow’s products and how we develop best-in-class models and techniques
- ML operations - designing and building highly automated ML pipelines, reproducible processes, and an infrastructure designed for continuous improvement
- Performance measurement - defining the metrics used to objectively measure the performance of the ML systems, diagnosing and troubleshooting issues
- Annotation and labelling processes - owning the tooling and processes behind how training data gets labelled to ensure that we have what we need to drive high-quality AI systems
- Edge/device and camera hardware input - technical input into camera and edge-compute decisions as new hardware generations come online, not just building models to fit existing constraints
- Privacy and responsible AI - broader ownership of how personal data is handled in the recognition pipeline (face-blurring is an early example), especially given customer security/compliance expectations
- Vendor and third-party model evaluation - managing relationships with external model providers and annotation partners as reliance on them grows
- Set and maintain the technical strategy for Winnow’s core computer vision pipeline, including food-tracking, waste-stage classification, weight estimation, and food-identity recognition models and beyond
- Take broader ownership of all ML/AI activities in the company (e.g. production forecasting) once the food recognition accuracy targets are met
- Act as an advisor on LLM-based product features with a view to owning this area over time
- Guide model improvement work to maximise performance of the overall system, including fully utilising all hardware capabilities (multiple cameras), and edge and cloud computing capabilities
- Oversee the creation and upkeep of high-quality datasets and annotation guidelines that the team’s models are trained and validated against
- Drive improvements to ground-truth and label quality, including label aggregation approaches for cases with multiple or conflicting annotations
- Sponsor and conduct exploratory research into new techniques - for example, using vision-language models as a verification layer on top of existing classifiers - and make informed calls on when to invest further versus park an experiment
- Set out a clear future-state vision, and a step-by-step plan to deliver it, for stakeholders across the business
- Own the team’s technical workspace, including data repositories, pipelines, toolsets, dashboards, and a combination of cloud and on-prem infrastructure
- Work closely with Engineering to build systems that combine software and AI to deliver high accuracy reliably, and ship models into production, including on edge hardware
- Partner with Product to translate customer and operational needs into a model roadmap, and to explain model performance and limitations in plain terms
- Represent ML and Data Science in technical reviews and sign-offs for new model releases, RFCs and major pipeline changes
- Build the business cases for ML investment and work with the Head of Technology on budget allocation
- Strong hands-on background in computer vision - object detection/tracking, classification, or segmentation - and comfort going deep on model architecture and evaluation when needed
- Experience with data quality and annotation workflows, including label aggregation or multi-annotator agreement techniques
- Experience running ML work through a clear operating rhythm (e.g. OKRs, RFCs, sign-off processes) in a fast-moving product environment
- Track record of shipping models into real-world production systems, ideally including edge or resource-constrained deployment
- Proven experience leading a machine learning or data science team, including direct people management, ideally within computer vision or a related applied ML domain
- You lead by example technically, while knowing when to step back and let your team own a problem
- You care as much about the rigour behind a result (data quality, evaluation methodology) as the result itself
- You get a lot done and can motivate and lead a team to do the same
- You’re comfortable making prioritisation calls with incomplete information, and can explain the reasoning to both engineers and non-technical stakeholders
- Exposure to using large vision-language or foundation models as auxiliary tools within a production ML pipeline (e.g. verification, labelling assistance)
- Experience in software engineering practices: automated testing, version control and breaking down larger software systems
- A PhD or MSc in a relevant field
- A background in a data-scarce or noisy-label domain where rigorous experimentation and ground-truth quality make an outsized difference
- Experience working with camera/IoT hardware constraints, or in a company whose product runs on physical devices in the field
- Experience with graph databases and distributed computing systems (e.g. Spark)
Skills
Sign in to read and join the discussion.