About Us:
Outpost is building the backbone of freight. We’re reinventing how supply chain infrastructure works in America with carrier agnostic truck terminals. As a vertically integrated real estate, operations, and technology company, we acquire and operate mission-critical real estate across the country to serve the largest logistics providers in the world. Backed by $1B from Greenpoint Partners, we’re scaling and building the most valuable logistics network in the country.
We thrive on accountability, integrity, and a shared drive to raise the bar. If you’re excited to reshape the industry alongside a high-performance team with a championship mindset that executes relentlessly, welcome aboard.
Role Summary:
Our platform combines AI-powered gate automation, computer vision, and operational software to help logistics operators run smarter, faster facilities. We're a small, high-conviction team shipping real software that ends up in real yards, at real gates, moving real freight; and we're growing fast, with revenue set to grow 10X over the next 18 months.
As we onboard more customers, our computer vision system sees more camera layouts, identifier types, and edge cases than ever. We need someone to own accuracy end-to-end: measuring it, understanding why we get it wrong, and turning that into the labeled data that makes our models better. Today that's mostly measurement and curation. Once the pipeline matures and moves into maintenance mode, we expect this role to also contribute fixes to the product itself, not just flag issues for others to resolve.
Key Responsibilities:
Own tracking and reporting of CV accuracy metrics, per customer and per identifier type.
Investigate misclassifications and false negatives, categorize root causes, and identify patterns across customers and yards.
Curate, label, and prioritize datasets for model retraining, partnering closely with our ML and CV engineers.
Build and improve the continuous learning pipeline so new models ship weekly with minimal manual engineering effort.
Define functional acceptance criteria for CV accuracy per customer and track progress against them.
Translate accuracy findings into decisions the engineering team and customer-facing stakeholders can act on.
As the pipeline matures, expect to move from flagging issues to fixing them directly; building the labeling/preprocessing tooling, running retraining jobs, and owning fixes for the error patterns you find, not just reporting them.
What You Can Expect:
Direct ownership over the metric that decides whether our product works in the real world.
A small team that moves fast, argues in good faith, and trusts engineers to make decisions.
Real influence on what the ML team builds next; your findings drive the roadmap, not the other way around.
Problems grounded in the physical world: gates, cameras, trucks, yards.
Qualifications:
3+ years in a data quality, ML data engineering or applied ML role.
Experience working with computer vision or object detection systems in production.
Comfortable writing Python for data analysis, pipeline automation, and dataset tooling.
Strong analytical rigor, comfortable digging into large volumes of imagery/data to find patterns, not just running a script and reporting a number.
Experience with dataset annotation/labeling tools and workflows (Roboflow, Labelbox, CVAT, or similar).
Strong communication skills in English — you write clearly and engage well async.
Preferred Qualifications:
Experience with continuous learning or active learning pipelines for production ML systems.
Familiarity with OCR systems and identifier recognition (plates, container numbers, etc.).
Experience partnering with customer success or support teams on quality metrics.
Background in QA/test engineering for ML systems.
Experience with Roboflow specifically.
Our Stack:
Python · Roboflow · VLM/OCR pipelines · GCP (GCS) · PostgreSQL · Snowflake · Node.js/TypeScript
Outpost is an Equal Opportunity Employer and Prohibits Discrimination of Any Kind.
