Data Scientist

📋 Description

• • Shape the future of responsible marketing at Zefr, the global leader in brand-safety technology for walled-garden social platforms. As a Data Scientist in our Tel Aviv hub, you will architect and deploy cutting-edge NLP and computer-vision models that process tens of millions of social posts every single day, ensuring that brands appear alongside content that is safe, relevant, and aligned with industry standards.
• • Own the full algorithmic lifecycle—from literature review and prototyping to production-grade deployment—using state-of-the-art transformer architectures, large language models (LLMs), and vision-language models (VLMs). You will design compound AI systems that fuse text, image, and video signals to deliver the most accurate and scalable content-classification service on the market.
• • Build and maintain massive-scale vector-search infrastructure that indexes hundreds of millions of content embeddings. By leveraging semantic and image-similarity search, you will enable real-time retrieval of relevant data, powering downstream classifiers and recommendation engines that operate at sub-second latency.
• • Balance cost, speed, and quality when selecting model families, quantization strategies, and cloud resources. You will run rigorous A/B tests, monitor drift, and iterate continuously so that every model release improves precision and recall without inflating compute spend.
• • Collaborate daily with cross-functional squads of ML engineers, backend developers, product managers, and policy experts. Together you will translate vague brand-safety requirements into concrete model objectives, define labeling schemas, and establish evaluation benchmarks that satisfy both clients and global regulators.
• • Contribute to our real-time data pipelines built on Apache Kafka, Snowflake, and Ray. You will write production-grade Python and SQL, orchestrate workflows in Airflow, and expose model endpoints via FastAPI so that downstream services—from dashboards to ad-servers—receive fresh inferences within minutes of ingestion.
• • Champion reproducible science. Using Weights & Biases, VertexAI, and GitHub Actions, you will version datasets, containerize experiments, and automate CI/CD so that every result is traceable, every model is auditable, and every rollback is painless.
• • Stay on the bleeding edge. You will monitor arXiv, attend top-tier conferences (NeurIPS, CVPR, ACL), and run internal reading groups to ensure Zefr adopts breakthrough techniques in few-shot learning, multimodal fusion, and efficient inference before the competition even hears about them.
• • Mentor junior scientists and data analysts, fostering a culture of curiosity, rigor, and psychological safety. Your code reviews, design docs, and lunch-and-learn sessions will raise the bar for scientific excellence across the entire R&D organization.
• • Finally, see your work in action. When global brands run campaigns on YouTube, Meta, TikTok, or Snap, your models decide—in real time—which videos are safe, which memes are on-brand, and which trends are worth riding. The impact is immediate, measurable, and visible to millions of users worldwide.