AI Engineer, Computer Vision

Mill•San Bruno, CA

About The Position

Mill is a waste prevention technology company focused on eliminating food waste through smart systems and infrastructure for homes, businesses, and municipalities. They transform food scraps into valuable resources like chicken feed. The company is launching Mill Commercial, an end-to-end solution for managing, understanding, and preventing food waste in commercial environments. This role involves working on the AI core of Mill Commercial, specifically the computer vision and agentic systems that process food waste streams into operational intelligence for commercial kitchens. The position integrates a camera and onboard compute into food recyclers, using edge models to identify, classify, and quantify food scraps. The vision pipeline then translates this data into procurement and operational guidance for large food service operators. The AI Engineer will join a small AI team, responsible for building the data and training pipeline for edge CV models, designing cloud-side evaluation processes, and developing agentic, LLM-driven product features that provide customer-facing insights and recommendations. This is a hands-on senior individual contributor role requiring expertise in fine-tuning segmentation models, prompting VLMs, and integrating agents into product features.

Requirements

Strong fundamentals in computer vision and deep learning — segmentation, detection, classification, tracking. You understand the architectures well enough to make informed choices.
Fluency with modern ML approaches — VLMs, LLMs, foundation models, and agentic systems — alongside classical deep learning. You know when to fine-tune a ConvNet, when to prompt a VLM, and when to wire up an agent, and you understand the practical realities of putting any of them into a product.
Experience building ML training pipelines and data annotation systems at scale.
Experience evaluating ML models rigorously — designing metrics, building the eval harness, and using results to drive product decisions rather than just publish a number.
Proficiency with cloud ML infrastructure (AWS or equivalent) — you've managed training jobs, data pipelines, and experiment workflows in production.
Familiarity with cloud-to-edge model deployment.
Clear, direct communication — you can explain tradeoffs to non-technical stakeholders, push back honestly when you disagree, and write docs that others can follow.
Genuine interest in applying AI to food waste reduction and sustainability. This is a mission-driven product and we want people who care about the mission.
Software skills: Python, PyTorch, OpenCV. Strong familiarity with MLOps on AWS infrastructure. Experience with LLM and agent frameworks. Google Cloud / Gemini experience is a plus.

Nice To Haves

Experience with video understanding (temporal consistency, tracking, video segmentation)
Experience with foundation models for data annotation
Experience with MLOps tooling (Weights & Biases, MLflow, SageMaker, or equivalents)
Experience shipping LLM- or agent-powered features in a consumer or B2B product
Hardware / IoT product experience, particularly with computer vision and cameras for embedded systems

Responsibilities

Build and manage the end-to-end ML training pipeline: data ingestion from deployed kitchen units, ground truth generation, annotation tooling (including foundation-model-assisted labeling), training, evaluation, and retraining cycles.
Train and evaluate segmentation, classification, and mass-estimation models for the Mill Commercial camera pipeline — from prompting foundation models to fine-tuning ConvNets and VLMs.
Build the cloud-side evaluation harness that tells us how our shipped edge models are actually performing in the field — automated, reproducible, and aligned to product accuracy targets across food types, kitchen environments, and deployment configurations.
Own MLOps: reproducible training, experiment tracking, model versioning, and automated evaluation against product-defined accuracy targets.
Export and validate models for deployment to edge devices, working closely with the edge team on optimization, quantization, and integration.
Help design and build the LLM- and agent-powered product features that consume waste characterization data and turn it into customer-facing recommendations — purchasing suggestions, anomaly explanations, operational nudges. Define how agents call tools, ground in customer data, and stay reliable in production.
Analyze failure cases systematically — unfamiliar food classes, novel kitchen environments, challenging lighting and clutter conditions — and drive the data and modeling decisions that close accuracy gaps.