I am interested in building AI systems that are useful, adaptive, and reliable in real-world settings.
My current focus is on post-training, reward modeling, reinforcement learning, and agentic AI systems. I am especially interested in how learning from feedback, preferences, and verifiable outcomes can make models more steerable and effective after pretraining.
Previously, my work has spanned natural language processing, information extraction, knowledge representation, fraud detection, and production machine learning systems. I care about both the modeling side and the systems side: how models are trained, evaluated, deployed, monitored, and improved in practice.
Current Interests
- Post-training and reinforcement learning for language models
- Reward modeling and preference learning
- AI agents and autonomous workflows
- Evaluation of LLM behavior and reliability
- Applied ML systems for high-stakes domains
Published Work
Rajat Patel. (2023). InterosML @ Causal News Corpus 2023: Understanding Causal Relationships: Supervised Contrastive Learning for Event Classification
Rajat Patel, Francis Ferraro. (2020). On the Complementary Nature of Knowledge Graph Embedding, Fine Grain Entity Types, and Language Modeling
Rajat Patel. (2020). Jointly Learning Knowledge Graph Embeddings, Fine Grain Entity Types and Language Models (Master thesis)
Writing
I write about reinforcement learning, post-training, LLM systems, and applied machine learning. Blog →