On top of my engineering work, I owned our computer vision product, reporting to the founder:

• Owned the roadmap, go-to-market and business model, steering the developers building it
• Scaled it to ~20 large accounts at top-5 distributors in six European countries, turning customer needs into product decisions
• Built a repeatable playbook to onboard customers, keep their models current, and help them cut manual testing and waste

OneThird predicts the shelf life and quality of fresh produce without damaging it. For strawberries, our largest category, I led the technical roadmap and built the data science and machine learning behind it:

• Automated shelf-life and Brix prediction from NIR spectra, metadata and images: from manually run algorithms to continuously retrained deep learning pipelines (custom 1D CNNs, transformers) that perform well for every customer
• Made noisy, subjective shelf-life labels more objective across customers, while keeping them aligned with each customer's own judgement
• Built the computer vision system from an early prototype to fine-grained models trained on 1M+ images, assessing size, colour and defects (white shoulders, wet spots, mold) to each customer's spec; a quality check now takes ~5 seconds instead of minutes
• Replaced Power BI with our own analytics stack for customer dashboards (Postgres, Cube.js, React)

After almost two years I left to build my own things full-time. I still work with OneThird on projects through Jacobian Systems.