Character.AI
Software Engineer, Applied ML (Discovery, Recommendation & Search)
About this role
Build and optimize ML infrastructure for Character.AI's discovery, recommendation, and search systems. You'll design end-to-end ML pipelines, implement serving infrastructure, and collaborate with product and data teams to power AI-generated content features on a rapidly growing consumer platform.
What you'll do
- Design and implement applied ML models and infrastructure for recommendation, ranking, and search systems
- Build ML backend systems powering discovery surfaces and AI-generated content formats
- Develop data pipelines, model training, and serving infrastructure
- Optimize model performance and infrastructure efficiency
- Collaborate cross-functionally with product, data science, and data platform teams
- Support and optimize existing ML systems
What they're looking for
- ML frameworks (PyTorch, TensorFlow)
- RESTful and gRPC web services
- Cloud infrastructure (GCP, AWS, or Azure)
- Modern typed programming languages
- CI/CD pipelines and automated testing
- Production ML systems and optimization
- Vector databases and feature storage
- End-to-end ML pipeline development
Opens the application — the Jobs AI extension fills it for you. Set up autofill
Opens the official application on the employer’s site. No login required.
Character.AI
Character.AI builds an interactive entertainment platform powered by AI characters that serves millions of users monthly, enabling multimodal interactions through advanced video, image, and language generation. The company is hiring research engineers, backend software engineers, ML specialists, and monetization experts to develop core infrastructure, scaling AI safety systems, and building reliable payment platforms.
View all jobs at Character.AILikely interview questions
- Walk us through an end-to-end ML feature you shipped to production—from data pipeline design through model serving. What were the main bottlenecks?
- Describe your experience optimizing ML models for production serving. How have you handled trade-offs between model accuracy and latency/cost?