CognitionSan Francisco, CABuilds core infrastructure for AI software agents in Devin and Windsurf, focusing on long-horizon task execution, tool use, planning, and reliability at scale. Requires strong Python, systems engineering depth, and AI curiosity; onsite in San Francisco.
Deploys state-of-the-art ML models and fine-tunes LLMs in production to enhance platform integrity and safety against adversarial threats. Requires Master's/PhD, deep learning expertise, PyTorch/TensorFlow proficiency.
Designs and runs experiments to improve oversight of increasingly capable AI models, including model training, evaluation, and deployment of practical systems. Analyzes failures and develops techniques to train more aligned models using oversight signals.
Builds and trains large-scale multimodal agentic models involving reasoning, planning, coding, and tool calling. Requires strong ML foundations, PyTorch expertise, and experience with distributed training on massive datasets.
Build benchmarks, datasets, and evaluation systems to measure and improve AI model quality for fraud, identity, and risk judgment tasks. Collaborate across research, engineering, and product to drive rigorous experimentation and iteration in high-stakes environments.
Research Engineer designs evaluations, studies model failures, and builds research loops to improve AI agents for high-stakes fraud detection and judgment tasks. Requires ML training experience, experimental rigor, and strong engineering skills in adversarial environments.