E
Senior Data Scientist
Sin especificarTiempo completoData ScienceEnterprise Software
Own the data science lifecycle for AI- and LLM-powered fraud and risk detection products, including modeling, prompt design, evaluation, experimentation, and production performance.
Responsabilidades
- Design, build, and evaluate LLM-based and classical machine learning models for classification, entity extraction, and risk scoring
- Lead prompt design and iteration, including model configurations, fallback strategies, cost, latency, and quality tradeoffs
- Apply statistical judgment to noisy and adversarial data
- Partner with engineers to deploy models in real-time streaming and large-scale batch systems
- Shape evaluation tooling, LLM observability, and model-performance monitoring
- Define infrastructure requirements that support rapid experimentation and production reliability
- Own data-quality and product metrics including precision, recall, coverage, latency, and cost efficiency
- Build measurement frameworks for offline experiments and production systems
- Analyze production data to identify labeling gaps, false positives, and new detection opportunities
- Drive A/B testing, shadow deployments, and offline evaluation using measurable outcomes
Requisitos
- At least 5 years of applied data science or machine learning experience, including production model deployment
- Hands-on experience with LLM prompt engineering and evaluation
- Strong Python skills
- Experience with large-scale data processing and data-intensive applications
- Understanding of modern data architectures and cloud-native systems; AWS experience is preferred
- Ability to work with noisy, adversarial, or imperfectly labeled data and develop labels or heuristics
- Strong communication and stakeholder-management skills
Se valora
- Experience with Apache Spark, Kafka, Kubernetes, Docker, EMR, Airflow, Iceberg, or Delta Lake
- Experience with Terraform, CI/CD, and observability platforms
- Familiarity with ML platforms, vector databases, AI evaluation frameworks, or MLOps