About the role
About the Company At Twenty, we're taking on one of the most critical challenges of our time: defending democracies in the digital age. We develop revolutionary technologies that operate at the intersection of the cyber and electromagnetic domains, where the speed of operations exceeds human sensing and complexity transcends conventional boundaries. Our team doesn't just solve problems – we deliver game-changing outcomes that directly impact national security. We're pragmatic optimists who understand that while our mission of protecting America and its allies is challenging, success is possible. Role Summary You’ll build and ship language-model-powered systems that strengthen Twenty’s mission-critical cyber capabilities for U.S. national security. You’ll own the end-to-end workflow—from curating specialized datasets and post-training models to deploying reliable inference and retrieval systems in production. You’ll partner closely with product and engineering to translate real operational needs into high-performing AI features, operating across cloud and on-premises environments where speed, correctness, and security matter.
Who You Are You’re motivated by real-world outcomes and want your work to directly impact national security missions. You care about rigor: clean data, measurable evaluation, and repeatable experiments beat demos. You balance research curiosity with product instincts—you ship, observe, iterate, and harden. You’re comfortable working across cloud and on-premises constraints and adapting to the environment. You communicate clearly with engineers and non-ML partners, and you write documentation people use. You think in systems: models, retrieval, infrastructure, and feedback loops all have to work together. You thrive in fast-moving teams with high standards, direct feedback, and high ownership. What You’ll Do Create, clean, and maintain high-quality training and evaluation datasets for specialized AI use cases. Fine-tune language models (small specialized through medium foundation models) for mission needs. Implement post-training and alignment approaches to improve task performance and reliability. Build retrieval-augmented generation (RAG) systems that ground model outputs in external knowledge. Develop and optimize model serving infrastructure for production deployments. Design evaluation frameworks and test harnesses to measure quality, latency, and regressions. Integrate AI capabilities into applications and workflows using modern orchestration frameworks. Collaborate with cross-functional partners to identify high-leverage use cases and deliver solutions. Produce clear technical documentation for models, datasets, and operational processes. Must Have You have 4+ years of professional software development experience building and supporting ML/AI-enabled applications. You have strong Python skills and deep learning experience with PyTorch, TensorFlow, or JAX. You have hands-on experience with LLM post-training methods (e.g., continued pre-training, SFT, RLHF, DPO, PPO, GRPO). You have experience curating, cleaning, and preprocessing datasets for training and evaluation. You have working knowledge of relational, graph, and vector database concepts. You have experience designing or using evaluation metrics and testing procedures for LLMs and agents. You have experience integrating LLM/agent systems using frameworks like Pydantic-AI, LangChain/LangGraph, or CrewAI. You have a Bachelor’s degree in Computer Science, Software Engineering, or a related field (or equivalent practical experience).