- Design, develop, and maintain scalable data pipelines to support ingestion, transformation, and delivery into centralized feature stores, model-training workflows, and real-time inference services. - Build and optimize workflows for extracting, storing, and retrieving semantic representations of unstructured data to enable advanced search and retrieval patterns. - Architect and implement lightweight analytics and dashboarding solutions that deliver natural language query experience and AI-backed insights. - Define and execute processes for managing prompt engineering techniques, orchestration flows, and model fine-tuning routines to power conversational interfaces. - Oversee vector data stores and develop efficient indexing methodologies to support retrieval-augmented generation (RAG) workflows. - Partner with data stakeholders to gather requirements for language-model initiatives and translate into scalable solutions. - Create and maintain comprehensive documentation for all data processes, workflows and model deployment routines. - Should be willing to stay informed and learn emerging methodologies in data engineering, MLOps and LLM operations.

• Design, build, and maintain backend APIs and services supporting credit risk and fraud decisioning • Deploy and operate machine learning models, scorecards, and business rules in production • Own system reliability, performance, and observability for underwriting and fraud services • Debug and resolve production issues affecting decision accuracy or service availability • Partner closely with data scientists to productionize models and ensure safe, repeatable deployments • Improve system architecture to support scalability, fault tolerance, and operational robustness • Implement monitoring and alerting for model behavior, decision outcomes, and system health • Collaborate with Credit Risk Strategy, Risk Modeling, Tech, and Operations teams to translate strategy into live decisions • Develop deep expertise in Katapult’s credit risk and fraud infrastructure

This description is a summary of our understanding of the job description. Click on 'Apply' button to find out more. Role Description The Content Understanding team builds machine learning models that "read" Pinterest content—images, text, and video—to produce high-quality semantic signals (e.g., embeddings, localization, quality/safety labels). These signals power relevance and retrieval for Homefeed, Search, Related Pins, and Ads, and also support integrity use cases like spam and low-quality detection. We work end-to-end: from data and labeling strategy, to model training and evaluation, to low-latency serving and monitoring at Pinterest scale. The role is ideal for a senior modeler who also enjoys developing, productionizing models and leading technical direction across teams. - Lead modeling strategy for content understanding (vision, NLP, multimodal), including architecture selection, training approach, and evaluation methodology. - Design and ship production models that generate content signals such as embeddings and classifications used across multiple product surfaces. - Own the full ML lifecycle: data/labeling strategy (human labels + weak supervision), training pipelines, offline evaluation, online experimentation, deployment, and monitoring/retraining. - Partner with infra/platform teams to ensure scalable, reliable training/serving (latency, cost, observability, rollout safety). - Collaborate with signal-consuming teams (ranking, retrieval, integrity, ads) to define signal contracts, adoption patterns, and success metrics. - Provide technical leadership through design reviews, mentoring, and raising the quality bar for modeling and ML engineering practices. Qualifications - M.S/ PhD degree in Computer Science, Statistics or related field. - Significant industry experience building software and ML pipelines/systems, including technical leadership (project/tech lead or equivalent). - Strong proficiency in Python and at least one ML stack such as PyTorch / TensorFlow, plus solid software engineering fundamentals. - Proven experience training and deploying ML models to production, including model versioning, rollouts, monitoring, and retraining strategies. - Deep hands-on experience in content understanding domains, such as: - computer vision (classification, detection, representation learning), - NLP (text classification, entity/topic modeling), - multimodal / embedding models (e.g., transformer-based representations). - Experience working with large-scale datasets and distributed compute (e.g., Spark-like ecosystems, distributed training, GPU environments). - Strong applied skills in evaluation and experimentation: defining metrics, offline/online alignment, A/B testing, debugging regressions, and model quality analysis. - Demonstrated ability to influence across teams and drive ambiguous problem areas to measurable outcomes. Requirements - This position is not eligible for relocation assistance. - This role will need to be in the office for in-person collaboration 1-2 times/quarter and therefore can be situated anywhere in the country. Benefits - Visit our PinFlex page to learn more about our working model.

• Train, fine-tune, and evaluate machine learning models on real-world workflow and behavioral data • Build predictive models for task outcomes, productivity trends, capacity forecasting, and workflow optimization • Fine-tune large models and foundation models for domain-specific prediction, classification, and embedding tasks • Design and maintain feature pipelines, training loops, and evaluation frameworks • Work with engineers and product teams to integrate trained models into production systems • Monitor model performance and iterate using offline evaluation and live data feedback