Applied Machine Learning Engineer

About Cozmo

Cozmo serves as an AI-driven operating system for property claims management. The platform supports both front-end customer interactions and back-end operations, handling tasks from incident reporting during emergencies to claim submissions through systems like Xactimate and Cotality, contractor dispatch based on service-level agreements, acceptance tracking, and preparation of carrier-ready estimates from onsite photographs.

Our clients include restoration franchisors, third-party administrators, and claims adjusting companies aiming to adopt AI-based solutions in their operations. A leading restoration franchisor in the U.S. anchors our customer base.

Role Overview

In this role, you will transform extensive, complex claims data—including line items, quantities, pricing, room dimensions, and approved versus submitted entries—into a valuable data asset. You will develop and deploy predictive models utilizing this data, navigating challenges such as inconsistent report formats, schema variability across franchises, varied carrier response formats, and labels reflecting negotiated outcomes rather than absolute truth.

Key Responsibilities

• Develop the claims data platform end-to-end, including data ingestion pipelines, schema creation, feature engineering, data storage, and workflow orchestration to support team scalability.
• Deploy the initial estimation models in production and refine them based on real claim outcomes.
• Design the labeling and evaluation frameworks suited for negotiated-label data, prioritizing correct metric selection to ensure model effectiveness.
• Integrate model predictions into live claim processes, collaborating closely with the field operation teams during deployment phases.
• Create tools to enable a small ML team to operate efficiently, such as automated training workflows, experiment tracking, model serving, and monitoring systems.
• Expand modeling capabilities to include advanced tasks like scope prediction from images and audio transcripts, supplement detection, and leakage identification.

Candidate Requirements

• Between 2 and 6 years of experience delivering machine learning systems into production with demonstrated pipeline resilience.
• Strong expertise in data engineering using tools such as Python, SQL, dbt, Spark, or equivalents, with experience managing infrastructure capable of handling schema drift and irregular real-world data.
• Proficiency in applied machine learning on tabular and structured datasets, with practical judgment on model selection favoring robust, interpretable solutions like gradient boosting when appropriate.
• Understanding of label validation and leakage prevention, capable of diagnosing and correcting issues where models may learn incorrect patterns despite appearing successful in offline evaluation.
• Preference for rapid production deployment and incremental improvement over theoretical model perfection.
• Demonstrated portfolio or documentation evidencing systems or data platforms designed and managed, including code, open source projects, or detailed problem-solving accounts.
• Enthusiasm to engage closely with the claims domain, such as comprehending estimates and collaborating with adjusters, to understand pricing mechanisms.

Preferred Qualifications

• Experience with structured data extraction from documents, photographs, or call transcripts.
• Familiarity with insurance data, risk analysis, pricing models, or marketplace data domains.
• Skills in large language model engineering for data extraction and assisting claim agents.

Work Environment

The team operates primarily onsite in New York City, working nearly six days weekly with additional hours during live launch cycles. This high-intensity setting leverages dedication as a competitive edge over much larger incumbents.

Compensation

Offers include competitive market salary plus equity at a founding level, emphasizing long-term value accumulation corresponding to the data assets developed in this position.

Application Process

Applicants should submit three items: a demonstration of a significant data pipeline or machine learning system delivered (with accessible code or write-up), a brief foundational answer to a question about training models on negotiated claim labels, and details on preferred work location and earliest availability. All genuine applications receive responses within 48 hours, with outstanding answers to the second question guaranteeing interview consideration.