·
Researchers & Thinkers
Omar Khattab
Assistant Professor |MIT EECS & CSAIL
Creator of DSPy (32.9k stars) and ColBERT. Pioneered declarative AI programming and late-interaction retrieval. Stanford NLP PhD, former Databricks Research Scientist. SIGIR 2025 Best Paper for WARP.
6papers
GitHubokhatResearch Papers
6 papersarXiv
Recursive Language Models
2512.24601·2025-12-31
arXiv
WARP: An Efficient Engine for Multi-Vector Retrieval
2501.17788·2025-01-29
arXiv
Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs
2406.11695·2024-06-17
arXiv
DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines
2310.03714·2023-10-05
arXiv
ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction
2112.01488·2021-12-02
arXiv
ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT
2004.12832·2020-04-27
Biography
Deep Research
Omar Khattab is an Assistant Professor at MIT EECS & CSAIL (TIBCO Founders' Career Development Professor) and the creator of DSPy and ColBERT. He completed his CS PhD at Stanford NLP, advised by Matei Zaharia and Christopher Potts, supported by the Apple Scholars in AI/ML Fellowship. Before joining MIT in July 2025, he was a Research Scientist at Databricks. His research investigates how to program intelligent software systems that are partly specified in natural language, that process natural language at scale, and whose quality and cost can be optimized using language models. DSPy has exceeded 500,000 downloads per month and sparked applications at Google, Amazon, IBM, VMware, Databricks, Baidu, AliExpress, and dozens of startups. ColBERT has won five SIGIR paper awards since 2020, including the SIGIR 2025 Best Paper for the WARP retrieval engine.
Declarative AI ProgrammingLanguage Model Program OptimizationNeural Information RetrievalLate Interaction Retrieval (ColBERT)Multi-Vector SearchRecursive Language ModelsNatural Language ProgramsRetrieval-Augmented GenerationFoundation Model ProgrammingScalable IR Infrastructure
Timeline
11 Research11 total
2026
2025
2024
2023
2023-01Research
DSPy framework open-sourced under stanfordnlp on GitHub2022
Key Contributions
DSPy
The framework for programming -- not prompting -- language models. A declarative programming model for composing and automatically optimizing natural language programs using modular signatures, with optimizers like MIPRO and GEPA. 32,900+ stars, 500K+ monthly downloads.
ColBERT
Late interaction retrieval model that independently encodes queries and documents with BERT, then uses fine-grained MaxSim for efficient scoring. Foundational to modern neural IR with 1,500+ citations and five SIGIR paper awards.
ColBERTv2 & PLAID
Improved ColBERT with residual compression and the PLAID indexing engine, dramatically reducing storage and latency while preserving retrieval quality.
WARP
Efficient engine for multi-vector retrieval achieving 3x speedup over ColBERTv2/PLAID with dynamic similarity imputation and implicit decompression. Won SIGIR 2025 Best Paper Award.
Recursive Language Models (RLM)
General inference paradigm allowing LLMs to programmatically examine, decompose, and recursively call themselves over long inputs, processing contexts up to two orders of magnitude beyond model windows.
MIPRO Optimizer
Algorithm for optimizing instructions and few-shot demonstrations in multi-stage language model programs, achieving up to 13% accuracy improvement over baselines.
STORM
Knowledge curation system built on DSPy that automatically researches and writes Wikipedia-style articles from scratch. Published at NAACL 2024.
Notable Quotes
“
DSPy is basically a programming language for building AI systems. We ask you to write your system in normal Python, but express your AI steps in the form of DSPy signatures. If you do that well, we give you two things automatically.
“
We need higher-level programming languages. We have only one: DSPy.
“
One of the biggest challenges facing DSPy is the lack of competition. As far as I can tell, there's just no other serious programming model for general-purpose, declarative AI programming.
12 sources(click to expand)
Omar Khattab - Personal WebsiteOmar Khattab - MIT EECS Faculty Profileokhat (Omar Khattab) on GitHubDSPy: The framework for programming -- not prompting -- language models (GitHub)ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT (SIGIR 2020)DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines (ICLR 2024 Spotlight)WARP: An Efficient Engine for Multi-Vector Retrieval (SIGIR 2025 Best Paper)Recursive Language Models (arXiv 2025)MIPRO: Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs (EMNLP 2024)Omar Khattab PhD Thesis - Stanford Digital RepositoryOmar Khattab on Google ScholarMIT School of Engineering welcomes new faculty in 2024-25
Research generated March 19, 2026
Researchers & Thinkers/Omar Khattab
All Profiles