Mukul Singh

Mukul Singh

Director of Science at Microsoft Research

PROSE Group, Redmond, USA

Email GitHub LinkedIn Google Scholar

I am a Director of Science at Microsoft Research in the PROSE (Program Synthesis using Examples) group. My research focuses on Agentic AI, Large Language Models, AI for Code, and Program Synthesis.

My work powers features in GitHub Copilot, Excel Copilot, Visual Studio, and Power Query, impacting over 1 billion users monthly. I develop code generation systems using inference-based, agentic, and fine-tuned frameworks, and have pioneered research on diffusion models for structured code generation.

I completed my B.Tech in Computer Science from IIT Delhi with a Minor in Artificial Intelligence.

Research Interests

  • Large Language Models for Code
  • Agentic AI Systems
  • Program Synthesis
  • Diffusion Models for Structured Generation
  • Table Understanding & Formatting

News

  • Jun 2026 New article: Knowledge Workers Don't Need Frontier Models — They Need Smarter Routing — nano-routed SLMs deliver near-frontier quality at 75–90% lower cost
  • May 2026 New article: Your AI Agent Forgets Everything: Why Persistent Memory Changes the Game — +15.6% accuracy in low-resource languages with executable memory
  • May 2026 Paper accepted at ACL 2026: "Agentic Context Strategies for Multi-Format Document Understanding"
  • May 2026 Paper accepted at HCAIR 2026: "Training Emergent Joint Associations: A RL Approach to Creative Thinking in LMs"
  • Apr 2026 Paper accepted at FSE 2026 Industry: "AutoChecklist: Automated Checklist Refinement for LLM Judges"
  • Apr 2026 Promoted to Director of Science at Microsoft Research
  • May 2026 New article: MCP Is Not Enough: Why Code DSLs Will Replace JSON Tool Calling — 60 experiments show 76% token savings vs. JSON tool calling
  • May 2026 Serving as Area Chair for FSE 2026 and Senior Area Chair for EMNLP 2026
  • Dec 2025 Invited speaker at Dagstuhl Seminar on Diffusion Models for Code Generation
  • Dec 2025 Paper accepted at NeurIPS 2025: "Diffusion is a code repair operator and generator"
  • Dec 2025 Serving as Senior Area Chair for ACL 2026
  • Sep 2025 Serving as Senior Area Chair for EMNLP 2025, ACL 2025, and NAACL 2025
  • Aug 2025 Promoted to Senior Researcher at Microsoft Research
  • May 2025 Keynote speaker at AIWARE conference co-hosted with ICSE 2025 in Seoul, South Korea
  • Apr 2025 Invited guest at Dagstuhl Seminar on Automated Programming and Program Repair
  • Nov 2024 Two papers accepted at EMNLP 2024: RAR and One-to-many testing for code generation
  • Apr 2024 Received Best Paper Award at ICSE 2024

Talks

More videos on Microsoft Research

EmFORE: Learning Email Folder Classification Rules by Example

AAAI 2024 Demo · 5:37

CODEFUSION: A Pre-trained Diffusion Model for Code Generation

EMNLP 2023 Oral Presentation · 6:18

CORNET: Learning Spreadsheet Formatting Rules by Example

Demo Presentation · 15:00

CORNET: Learning Table Formatting Rules By Example

VLDB 2023 · 4:25

Articles

Knowledge Workers Don't Need Frontier Models — They Need Smarter Routing

A nano-model router dispatching between GPT-5.5 and GPT-5.4 Mini reaches #2 on GDPVal with 75–90% cost reduction. Here's why SLMs + intelligent routing is the architecture that scales AI to knowledge workers.

AI Strategy SLM Routing Knowledge Workers Cost Optimization

Your AI Agent Forgets Everything: Why Persistent Memory Changes the Game

We gave LLMs persistent executable memory and measured +15.6% accuracy in low-resource languages, 33% fewer reasoning steps, and cross-model portability without retraining.

AI Agents Memory Code Generation LLM

MCP Is Not Enough: Why Code DSLs Will Replace JSON Tool Calling

We ran 60 experiments and found Code DSLs use 76% fewer tokens, 48% fewer turns, and 45% lower latency at identical quality compared to JSON tool calling.

AI Agents MCP Code Generation LLM

Publications

Full list on Google Scholar

2026
2025
2024
2023
2022
2021
2020

Patents

  • AI Engine Retrieving Over Typed Documentation Graphs to Generate Code Mukul Singh
  • Systems and Methods for Executable Directory-Structured Memory in Large Language Models Mukul Singh
  • Execution Guided Within-Prompt Search With Large Language Models Mukul Singh, Ashish Tiwari, Vu Le, Sumit Gulwani, Gust Verbruggen
  • System and Method for Intelligent Chart Design Recommendations in Spreadsheet Applications Using Large Language Models Mukul Singh, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Multistep retrieval for grammar based code generation Mukul Singh, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Dynamic Refinement of Language Model Evaluation Checklists Mukul Singh, Mansi Uniyal, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Tabular Data Augmentation Through Semantic Text Analysis and Labeling Mukul Singh, Ananya Singha, Vu Le, Sumit Gulwani
  • Recognizing tables in unstructured text Mukul Singh, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Vision chain-of-thought reasoning for multimodal language models Mukul Singh, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Code generation system using pre-trained diffusion model Mukul Singh, Jose Cambronero, Carina Negreanu, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Database rule generation for tabular task automation suggestion and ranking Mukul Singh, Jose Cambronero, Carina Negreanu, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Learning Syntactic and Semantic String Repairs Mukul Singh, Jose Cambronero, Carina Negreanu, Vu Le, Sumit Gulwani, Gust Verbruggen
  • Target Property Selection Techniques for Learning What to Teach Language Models for Code Generation Anirudh Khatry, Mukul Singh, Gust Verbruggen, Sumit Gulwani, Vu Le
  • Automated recommendation and ranking of personalized user interface actions in task and data automation platforms Saksham Gupta, Mukul Singh, Gust Verbruggen, Sumit Gulwani, Vu Le
  • Automated system for battery health and capacity prediction using charge and voltage statistics Mukul Singh, Bijaya Ketan Panigrahi, Akhil Garg

Experience

Director of Science

Apr 2026 – Present

Microsoft Research, PROSE Group · Redmond, USA

Leading research on Agentic AI, code generation, and program synthesis. Work powers GitHub Copilot, Excel Copilot, and Visual Studio.

Researcher

Jul 2022 – Mar 2026

Microsoft Research, PROSE Group · Redmond, USA

Research on code generation, diffusion models, table understanding, and UI automation. Work powers GitHub Copilot, Excel Copilot, and Visual Studio.

Research Intern

Jun 2021 – May 2022

Microsoft Research · Redmond, USA (Remote)

Developed neurosymbolic AI for automatic spreadsheet formatting. Led to publications in VLDB and SIGMOD.

AI Research Intern

May 2022 – Jul 2022

Meta (Facebook) · London, UK

Designed auto-encoder based anomaly detection for manufacturing defect identification.

AI Research Intern

May 2021 – Jul 2021

American Express · Gurgaon, India

Deep learning for email subject line optimization and open rate prediction.

Research Intern

Dec 2019 – Jun 2020

Robert Bosch GmbH · Bangalore, India

Vehicle reaction time estimation using driving pattern data for chassis health evaluation.

Education

B.Tech in Computer Science and Engineering

2018 – 2022

Indian Institute of Technology Delhi

CGPA: 9.2/10.0

Minor in Artificial Intelligence

2018 – 2022

Indian Institute of Technology Delhi

CGPA: 9.6/10.0

Academic Service

Senior Area Chair

EMNLP 2026, ACL ARR 2026, ACL 2026, EMNLP 2025, ACL 2025, NAACL 2025

Area Chair

FSE 2026, EMNLP 2024, ICSE 2024, CIKM 2024, ACL 2024

Program Committee / Reviewer

NeurIPS 2026, ICLR 2026, COLM 2026, IJCAI-ECAI 2026, ACL 2026, ACM CAIS 2026, ACL, EMNLP, AAAI, NeurIPS

Journal Reviewer

IEEE TIV, IEEE ITS, SIGMOD, OOPSLA

Awards & Achievements

  • Dagstuhl Invited Guest – Automated Programming and Program Repair Seminar, Germany (2025)
  • ICSE Keynote Speaker – AIWARE conference co-hosted with ICSE 2025, Seoul, South Korea (2025)
  • ICSE Best Paper Award – International Conference on Software Engineering, USA (2024)
  • SMU Invited Talk – Research keynote at Singapore Management University (2024)
  • NeurIPS Best Reviewer Award – Table Representation Learning Workshop (2023)
  • VLDB Invited Speaker – Tabular Data Analysis Workshop, Canada (2023)
  • Sustainability Research & Leadership Award – AI for vehicle automation, Tata Power (2022)
  • Young Researcher Award – Bachelor thesis, NSS & Govt. of India (2021)
  • TIDE Fellowship – Tech Entrepreneurship Award & Incubation by MeitY, Govt. of India (2020)
  • Merit Scholarship – Full merit scholarship for scholastic performance, IIT Delhi (2018)
  • KVPY Fellowship – Kishore Vaigyanik Protsahan Yojana, IISc Bangalore & Govt. of India (2017)
  • InPhO Merit Certificate – Outstanding performance in International Physics Olympiad (2016)