ML Infrastructure Engineer

About the Role

You will own the pipeline that keeps our AI running as we scale: data from every site, into training, into models running back on site.

We are growing from a handful of live deployments to many sites, each with more cameras and robots. Our CV engineers build the models. Your job is to make sure lakhs of images arrive organized, get labelled and checked, feed automatic retraining, and come back as new models running on edge devices, with full visibility of which version runs where.

This is a platform and automation role. You will be our first dedicated hire for it, which means real ownership of architecture decisions from day one.

Who You'll Work With

You report to the CEO and work daily with:TeamWhat they doHow you work togetherCV / ML EngineersBuild and improve modelsThey define what data and evaluations they need; you make it flow and automate trainingAnnotation & QC teamLabel data and check qualityThey set label rules; you build the tooling and automated quality checksDashboard DeveloperBuilds the client dashboardYou define the APIs and data contracts that feed itDeployment Robotics EngineersInstall and commission robots on siteThey own the hardware; you push and monitor models on itWhat You'll Do

• Data ingestion: Build the pipeline that moves images and sensor data from every site to the cloud, organized by client, site and device, with nothing lost or duplicated
• Datasets: Set up dataset storage and versioning so any model can be traced back to the exact data it was trained on
• Annotation at scale: Automate the annotation workflow, including pre-labelling, mislabel detection and QC checks, so labelling keeps pace with incoming data
• Automated retraining: Turn training from a manual run into a pipeline that triggers on new labelled data and reports evaluation results for the CV team to approve
• Edge deployment: Package approved models and roll them out to on-site GPU devices in stages, with rollback and a clear record of which version runs where
• Monitoring: Set up alerts for failed data uploads, offline devices and model performance drops, so we know before a client does
• Dashboard data: Deliver operational data from each site to the dashboard through clean, documented APIs
• Cloud: Run our AWS setup (storage, compute, training jobs) with sensible costs
• Documentation: Write down the architecture, retraining process and deployment steps so the system doesn't depend on one personWhat You Bring
• 2–5 years building data pipelines, ML pipelines or MLOps systems that run in production
• Strong Python and SQL, with experience handling large datasets (lakhs of images or records)
• Hands-on AWS (or GCP/Azure): object storage, compute, and running jobs in the cloud
• Experience with a workflow or pipeline tool such as Airflow, Prefect, Dagster or Kubeflow
• Docker, Git and basic CI/CD
• Working knowledge of computer vision data: image datasets, annotation formats (COCO, YOLO), and how training and evaluation work
• Experience deploying models for inference, in the cloud or on edge devices
• You automate repetitive work by default and document what you buildYou don't need to tick every box. If you have built a pipeline that runs on its own and kept it running, we want to talk to you.

Nice to Have

• Edge GPU deployment (NVIDIA Jetson, TensorRT, ONNX)
• Dataset versioning or experiment tracking tools (DVC, MLflow, Weights & Biases)
• Annotation tools such as CVAT or Label Studio
• Streaming or IoT messaging (Kafka, MQTT)
• Infrastructure as Code (Terraform) and Kubernetes
• Monitoring tools (Prometheus, Grafana, CloudWatch)
• Keeping data from multiple clients cleanly separated
• Startup experience, or exposure to robotics, industrial automation or recyclingWhat Success Looks Like
• First 3 months: Data from every active site lands in the cloud automatically, organized by client and site
• First 6 months: New labelled data triggers retraining and evaluation without anyone starting it by hand
• First 12 months: Approved models roll out to all sites in days, we can see which version runs where, and site issues are caught by alertsWhy Join Us
• Build the platform from scratch and own its architecture, instead of maintaining someone else's
• Work across cloud, edge GPUs and real robots, a combination few roles offer this early in a career
• See your systems sort real waste at real industrial sites
• Direct access to the CEO and a direct say in how the company scales

HR@wastefullinsights.com