Staff ML Engineer leading Reddit’s ads retrieval modeling: designing candidate-generation and two-tower retrieval systems, running experiments, and setting technical direction while mentoring engineers on the Retrieval & Targeting team.
Leads ML forecasting and scenario modeling for Reddit's ads platform, building models that predict campaign outcomes under different budgets, bids, audiences, and creative performance. Owns modeling direction, production ML systems, and technical leadership across advertiser planning and recommendations.
Staff ML Engineer on Reddit's Ads Creative Effectiveness team building GenAI products for ad creative generation (image/video/copy), brand-safe editing pipelines, and LLM/VLM fine-tuning with RLHF/RLPF using ad performance metrics.
Staff Machine Learning Engineer building advanced ML models (DNNs, transformers) for Reddit's Ads Ranking system, focusing on app ads modeling, feature engineering, and large-scale recommendation systems using TensorFlow/PyTorch.
Senior Staff ML Systems Engineer leading technical strategy for Reddit's Ads ML Platform, focusing on feature development, training data, offline experimentation, and model iteration workflows using distributed systems like Spark, Flink, Kafka, and Kubernetes.
Staff ML Engineer leading Reddit's Ads Retrieval team, designing large-scale candidate generation models (two-tower, embeddings, ANN) using TensorFlow/PyTorch to improve ad relevance for 130M daily users.
Staff Machine Learning Engineer at Cresta building production-grade agentic AI systems for contact centers, including LLM-powered agent augmentation, evaluation frameworks, and RAG pipelines. Requires 7+ years ML production experience, deep LLM expertise, and leadership in architecting complex AI systems.
Staff ML Tech Lead Manager owning the machine learning stack for Anduril's Counter Intrusion team, designing and deploying multi-sensor computer vision models on edge devices using C++, Python, PyTorch, and TensorRT for real-time defense applications.
Staff Software Engineer on Google Workspace's Gmail Trust and Compliance Infrastructure, building batch processing and ML infrastructure to power GenAI features across Gmail, Docs, Drive, and other Workspace products while collaborating with Google DeepMind model builders.
Who we are At Twilio, we’re shaping the future of communications, all from the comfort of our homes. We deliver innovative solutions to hundreds of thousands of businesses and empower millions of developers worldwide…
At Braze, we have found our people. We’re a genuinely approachable, exceptionally kind, and intensely passionate crew. We seek to ignite that passion by setting high standards, championing teamwork, and creating…
Google's software engineers develop the next-generation technologies that change how billions of users connect, explore, and interact with information and one another. Our products need to handle information at…
At Affirm, we exist for the moments that matter—giving people a clear, predictable way to pay over time, with no hidden fees, no surprises, and no tradeoffs on what matters most. We are looking for a Staff Machine…
Designs and ships LLM safeguards (prompt injection, jailbreak, toxicity, sensitive data detection) in an agent harness, builds safety evals, fine-tunes detection models, and works with security teams. Uses Python and applied ML in production systems.
Discord has a highly engaged community of millions of daily active users who use the platform for many different reasons, but there’s one thing that nearly everyone does: play video games. Discord plays a uniquely…
Staff ML engineer on Google's Discover Personalization team building large-scale user models, neural retrieval, and ranking systems for Search using TensorFlow, C++, and advanced ML techniques like transformers and reinforcement learning.
Cookies?
A session cookie keeps you signed in. If you accept, we also use analytics cookies (Google Analytics and PostHog) to understand usage. Learn more
We couldn't check your fit for this role — add a CV to your profile to see it next time.