Omar Khattab is an assistant professor of electrical engineering and computer science at MIT, a principal investigator at CSAIL, and the creator of DSPy and ColBERT. His research makes AI systems more programmable, reliable, and adaptable as models, optimization methods, and application requirements change.
Khattab earned a computer science degree from Carnegie Mellon University in Qatar in 2019 and completed his Stanford doctorate in January 2025 under Christopher Potts and Matei Zaharia. His doctoral work received an Apple Scholars in AI/ML fellowship, and he interned at Apple in 2022. He worked as a research scientist at Databricks from June 2024 through June 2025 before joining MIT that July, where he holds the TIBCO Founders’ Career Development Professorship and leads the Open Source AI Systems laboratory.
During his first quarter at Stanford, Khattab began developing ColBERT with Zaharia. Its late-interaction retrieval architecture encodes documents in advance while preserving comparisons between individual query and document tokens, making detailed neural search practical across large collections. The resulting open-source ColBERT project helped establish an alternative to compressing every document into a single embedding.
In 2023, Khattab and collaborators introduced DSPy, a framework for specifying language-model tasks, composing programs, measuring results, and optimizing their behavior. Its declarative signatures separate what an application must accomplish from model-specific prompts, reasoning strategies, and formatting conventions.
- Durable AI engineering: Khattab distinguishes increasingly capable general intelligence from dependable software. An application still requires explicit objectives, evaluation criteria, tools, privacy boundaries, and control flow. His approach to enduring AI systems keeps those requirements stable while allowing models and optimization strategies to change.
- Reflective optimization with GEPA: Khattab co-developed GEPA, which analyzes execution traces, diagnoses errors in natural language, proposes revisions, and evolves promising alternatives. He has also highlighted Lakshya Agrawal’s related work extending reflective optimization to code and other artifacts.
- Recursive Language Models: With Alex Zhang and Tim Kraska, Khattab co-authored Recursive Language Models, an approach that treats exceptionally long inputs as an external environment a model can inspect, partition, and process recursively.
At MIT, Khattab’s laboratory applies these ideas to last-mile AI adaptation: helping general-purpose systems acquire the specialized knowledge and capabilities needed for particular documents, tasks, and domains.
