Principal Computer Vision Engineer

Instacart is a Flex First team

There’s no one-size fits all approach to how we do our best work. Our employees have the flexibility to choose where they do their best work—whether it’s from home, an office, or your favorite coffee shop—while staying connected and building community through regular in-person events. Learn more about our flexible approach to where we work.

Overview

At Arpalus, recently acquired by Instacart, we build AI-powered computer vision and AR technology that turns a phone camera into a precise sensing tool - real-time
object detection, scanning, and spatial understanding shipped as a mobile product. Our tech runs on real devices, in real environments, under messy real-world conditions.

We're a small, fast-moving team and we're looking for a Principal Computer Vision Engineer who already works that way too.

You'll join our core AI team and help craft best-in-class perception systems thatpower our real-time retail analytics. You'll solve genuinely hard problems in computer
vision, bridging the gap between cutting-edge research and highly optimized production pipelines. You'll work hands-on, architecting solutions that seamlessly blend classic computer vision techniques with modern deep learning.

In this role, you'll partner closely with our mobile and backend teams to deliver unique capabilities. As a Principal Computer Vision Engineer, you'll help shape the technical strategy for the AI group, influence direction across our training and inference codebases, and ship high-quality, observable algorithms at scale.

We value clear communication, practical problem solving, and ownership. If you thrive in a fast-paced, evolving environment where you can roll up your sleeves, make
thoughtful tradeoffs, and see your work move key metrics in real time, you'll feel right at home here.

About the Job

• Architect and Own: Drive the evolution of our core computer vision pipelines, taking algorithms from initial research and training through to highly optimized production deployment.
• Blend Classic & AI: Design and implement robust classic computer vision algorithms alongside deep learning architectures to solve complex spatial, alignment, and tracking challenges.
• Model Development: Train, evaluate, and fine-tune state-of-the-art deep learning models for object detection, classification, and advanced segmentation.
• Optimize for Production: Build and optimize high-performance inference pipelines in Python and C++, leveraging optimization tools like TensorRT and CUDA to ensure low-latency execution.
• Technical Leadership: Drive technical excellence across the AI stack, evaluating new architectures, standardizing our training methodologies, and making pragmatic tradeoffs between accuracy, speed, and compute constraints.
• Deliver End-to-End: Handle technical design, clean architecture, implementation, observability, and iterative improvements of our perception engine.

Minimum Qualifications

• 6+ years of industry experience in computer vision, machine learning, or a related field.
• Deep expertise in classic computer vision techniques, including strong practical knowledge of optical flow, multi-view geometry, feature matching, and homography

estimation (e.g., using OpenCV).

• Strong foundational understanding of modern deep learning architectures, specifically for object detection, image classification, and semantic/instance segmentation.
• Production-level programming skills in Python.
• Extensive hands-on experience with PyTorch and building scalable model training andevaluation pipelines.
• Proven track record of optimizing complex algorithms for deployment in resource-constrained or real-time environments.

Preferred Qualifications

• Familiarity with state-of-the-art segmentation frameworks and zero-shot/foundation models (e.g., SAM, Mask2Former, Grounding DINO).
• Experience working alongside mobile teams to integrate computer vision models into native SDKs (iOS/Android).
• Experience translating computer vision algorithms into Apple frameworks and iOS.
• Background in retail tech or AR-based scanning applications.