Peeyush Kumar

Peeyush Kumar

AI Researcher · Builder · Social Innovator · Entrepreneur

Previously: Senior Researcher, Microsoft Research · Co-founder & CTO, Engooden Health
Ph.D. University of Washington Seattle · M.Tech & B.Tech IIT Madras

I am a systems & AI researcher building technology that bridges people, process, and infrastructure in regenerative ways. My work spans agentic orchestration, large language models, reinforcement learning, and operations research - applied to commerce, energy systems, agriculture, food supply chains, healthcare, and community economics. I focus on using large AI models to create equitable, resilient systems at the intersection of industry and society.

Scholar LinkedIn Website Substack

Research Projects

Building AI systems for real-world impact - from LLM reasoning models, LLM architectures to energy microgrids powering underserved communities that ground knowledge in domain knowledge.

AI reasoning and intent illustration

LLM Reasoning for Underspecified Intent

A Bayesian learning framework for intent tracking that models user goals as they evolve across interaction steps and execution states. Infers latent objectives, identifies missing constraints, and maps high-level intent into executable task structures. Achieved 15% improvement in intent prediction accuracy.

LLM Reasoning Bayesian Learning
Rubric learning for grader evolution

Rubric Learning for Grader Evolution for Fine-Tuning LLMs

A framework for constructing and iteratively refining graders that encode task-specific notions of quality, completeness, and appropriateness. Treats the grader as an adaptive object that improves through exposure to completed jobs and outcome signals. Produced 20% improvement in evaluation quality for fine-tuning datasets.

LLM Fine-Tuning Evaluation
Cultural diversity and AI evaluation

Evaluation Design for Long-Form LLM Generation for Culturally Sensitive Tasks

An ontology-driven evaluation framework for identifying implicit cultural bias in long-form model outputs. Specifies measurable criteria for outputs whose success depends on cultural sensitivity and contextual appropriateness. Achieved 20% better detection of implicit cultural bias relative to baseline methods.

LLM Evaluation Cultural Bias
OG-RAG: Ontology-Grounded Retrieval-Augmented Generation

OG-RAG: Ontology-Grounded Retrieval-Augmented Generation

A novel method that anchors LLM retrieval in domain-specific ontologies using hypergraph representations. Achieves 55% better fact recall and 40% improved correctness across four LLMs.

LLM RAG Knowledge Graphs EMNLP 2025
Paper Code
Solar panels and renewable energy

AI-Powered Microgrids for Energy Resilience & Equity

Collaborating with rural communities to deploy AI-optimized microgrids that reduce energy costs for low-income households while maximizing renewable energy usage through reinforcement learning.

Reinforcement Learning Energy Equity
Paper Blog Project
Community and energy infrastructure

Community Economics for AI-Powered Micro-Grids

Designing economic models and optimization frameworks for community-owned micro-grids, enabling equitable energy distribution and participatory decision-making.

Economics Optimization Community Systems
Paper Project
Wind turbines at sunset

Multi-Market Energy Optimization with Renewables

Reinforcement learning framework for optimizing energy dispatch across multiple markets, determining when to generate, store, or sell electricity from renewable sources.

RL Energy Markets Optimization
Paper Project
Precision agriculture field

DeepMC: Micro-Climate Prediction Framework

Multi-scale encoder-decoder deep learning framework for hyper-local weather prediction, enabling precision agriculture and renewable energy planning at the individual farm level.

Deep Learning Climate Agriculture KDD 2021 NeurIPS 2023 NeurIPS 2024
Paper Blog Project
Supply chain logistics

Re-Inventing the Food Supply Chain with IoT

Data-driven platform combining IoT sensors, molecular tags, and AI to reduce food loss, enable traceability, and optimize cold chain logistics from farm to table.

IoT Supply Chain Food Systems
Paper Project
Drone over farmland

Democratizing Data-Driven Agriculture

Making AI-powered agriculture accessible through affordable hardware, partnering with Land O'Lakes and the UN FAO to bring precision farming tools to smallholder farmers globally.

Affordable AI Hardware IEEE Micro
Paper Project
Drone flying over landscape

Visage: Timely Analytics for Drone Imagery

Edge-cloud system enabling real-time analytics on drone imagery for agricultural monitoring, environmental sensing, and infrastructure inspection.

Computer Vision Edge Computing MobiCom 2021
Paper Project
Network graph visualization

Privacy-Preserving Multi-Agent RL for Supply Chains

Multi-agent reinforcement learning framework for optimizing supply chain decisions while preserving privacy of participating entities through secure computation.

Multi-Agent RL Privacy Supply Chain
Paper
Healthcare technology

Engooden Health: Clinical AI Platform

Co-founded healthcare AI company processing clinical language to enable chronic care management, cohort identification, and evidence-based decision support at the point of care. Managing ~100K patients monthly.

NLP Healthcare Startup

Selected Publications

2025
OG-RAG: Ontology-Grounded Retrieval-Augmented Generation for Large Language Models
Kartik Sharma, Peeyush Kumar, Yunqing Li
EMNLP 2025
PDF Code
2024
AI-Powered Microgrids Facilitate Energy Resilience and Equity in Regional Communities
Peeyush Kumar et al.
MSR Technical Report, MSR-TR-2024-44
PDF
2023
Zero-Shot Micro-Climate Prediction with Deep Learning
Iman Deznabi, Peeyush Kumar, Madalina Fiterau
NeurIPS 2023 Workshop: Tackling Climate Change with Machine Learning
Privacy Preserving Multi-Agent Reinforcement Learning in Supply Chains
Ananta Mukherjee, Peeyush Kumar, Boling Yang, Nishanth Chandran, Divya Gupta
arXiv:2312.05686
PDF
Community Economics for AI Powered Micro-Grid
Peeyush Kumar
Microsoft Research, 2023
PDF
Multi-Market Energy Optimization with Renewables via Reinforcement Learning
Lucien Werner, Peeyush Kumar
arXiv:2306.08147
PDF
Reward Shaping via Diffusion Process in Reinforcement Learning
Peeyush Kumar
arXiv:2306.11885
PDF
2022
Re-Inventing the Food Supply Chain with IoT: A Data-Driven Solution to Reduce Food Loss
Vaishnavi Ranganathan, Peeyush Kumar, Upinder Kaur, Sarah H.Q. Li, Tusher Chakraborty, Ranveer Chandra
IEEE Internet of Things Magazine, Vol 5(1), 2022
PDF
Democratizing Data-Driven Agriculture Using Affordable Hardware
Ranveer Chandra, Manohar Swaminathan, Tusher Chakraborty, Jian Ding, Zerina Kapetanovic, Peeyush Kumar, Deepak Vasisht
IEEE Micro, Vol 42(1):69-77, 2022
PDF
General Sum Stochastic Games with Networked Information Flows
Sarah H.Q. Li, Lillian J. Ratliff, Peeyush Kumar
Gamification and Multi-Agent Solutions @ ICLR 2022
2021
Micro-Climate Prediction - Multi Scale Encoder-Decoder Based Deep Learning Framework
Peeyush Kumar, Ranveer Chandra, Chetan Bansal, Shivkumar Kalyanaraman, Tanuja Ganu, Michael Grant
ACM SIGKDD (KDD) 2021
PDF
Visage: Enabling Timely Analytics for Drone Imagery
Sagar Jha, Youjie Li, Shadi Noghabi, Vaishnavi Ranganathan, Peeyush Kumar, Andrew Nelson, Michael Toelle, Sudipta Sinha, Ranveer Chandra, Anirudh Badam
ACM MobiCom 2021
PDF
Affordable Artificial Intelligence - Augmenting Farmer Knowledge with AI
Peeyush Kumar, Andrew Nelson, Zerina Kapetanovic, Ranveer Chandra
FAO/ITU Digital Agriculture in Action, 2021
2018 & Earlier
Information Directed Policy Sampling for POMDPs with Parametric Uncertainty
Peeyush Kumar, Archis Ghate
INFORMS Advances in Service Science, Springer, 2018
PDF
Near Optimal Hamiltonian-Control and Learning via Chattering
Peeyush Kumar, Wolf Kohn, Zelda B. Zabinsky
arXiv:1703.06485, 2017
PDF
Multi-Timescale, Gradient Descent, Temporal Difference Learning with Linear Options
Peeyush Kumar, Doina Precup
arXiv:1703.06471, 2017
PDF
Option Discovery in Hierarchical Reinforcement Learning using Spatio-Temporal Clustering
Peeyush Kumar, Aravind S. Lakshminarayanan, Ramnandan Krishnamurthy, Balaraman Ravindran
ICML 2016 Workshop on Abstraction in RL
PDF
Spectral Clustering as Mapping to a Simplex
Peeyush Kumar, N. Narasimhan, Balaraman Ravindran
ICML 2013 Workshop on Spectral Learning
Abstraction in Reinforcement Learning in Terms of Metastability
Vimal Mathew, Peeyush Kumar, Balaraman Ravindran
10th European Workshop on Reinforcement Learning (EWRL), 2012

Full list on Google Scholar.

Patents

Mentorship

Mentored PhD, Masters, and undergraduate students across top universities. Many have gone on to faculty positions and leading research roles.

Sarah Li
PhD, University of Washington Seattle (2021)
Now: Assistant Professor, Georgia Tech
Upinder Kaur
PhD, Purdue University (2021–2023)
Now: Assistant Professor, Purdue
Jessica Quaye
UG, MIT (2021)
Now: PhD student, Harvard
Lucien Werner
PhD, Caltech (2021)
Alireza Sadeghi
PhD, University of Minnesota (2021)
Now: 3M
Christopher Cross
Masters, Stanford (2022–2023)
Boling Yang
PhD, University of Washington (2022–2023)
Yunqing Li
PhD, North Carolina State University (2023)
Youya Xia
PhD, Cornell (2021)
Kowshik Thopalli
PhD, Arizona State University (2020)
Now: Researcher, National Lab
Andalib Samandari
UG, Georgia State University (2023)
Kartik Sharma
PhD, Georgia Tech
Now: Microsoft
Emiliano Diaz
Yale
Sagnik Mukherjee
PhD, UIUC

Talks & Service

Selected Talks & Panels

  • Economic Anthropology in the Age of Generative AI - AAA Annual Meeting 2024
  • Economic Anthropology in the Age of Generative AI - SEA/SAW Joint Meeting 2024
  • Zero-Shot Learning - NeurIPS 2023
  • INFORMS Annual Meeting 2022
  • Cybersecurity, Water and Food Symposium - Virginia Tech 2022
  • DOE AI4ESP Panel - AI for Earth System Predictability 2021
  • ACM KDD - Conference on Knowledge Discovery & Data Mining 2021
  • UN FAO/ITU Digital Agriculture Solutions Forum - Asia-Pacific 2020
  • MIT Sloan Business School - Invited Presentation 2018
  • George Mason University - Invited Presentation 2018
  • ICML 2016
  • ISMP / RLDM 2015

Academic Service

  • UK BRAID (Bridging Responsible AI Divide) - Grants Reviewer 2024
  • Microsoft AI & Society Fellowship - Fund Reviewer 2024
  • Manufacturing & Service Operations Management Journal - Reviewer 2023
  • IISE Transactions - Reviewer 2023
  • ACM PACM IMWUT - Reviewer 2023
  • ACM AgSys Workshop - PC Member & Reviewer 2023
  • O2 VC Fund - Mentor & Reviewer
  • IEEE Trans. on Control of Network Systems - Reviewer
  • IEEE Internet of Things Magazine - Reviewer
  • UW Global Innovation Exchange - Competition Judge 2021
  • PhD Committee - Iman Deznabi, UMass

Teaching

  • Probability and Statistics (Graduate & Undergraduate) - University of Washington Seattle 2019
  • Engineering Economics (Graduate & Undergraduate) - University of Washington Seattle 2016
  • Biomechanical Modelling (Graduate) - IIT Madras 2013

Education & Awards

Education

  • Ph.D. in AI & Operations Research
    University of Washington Seattle
    2013 – 2018
    Thesis: "Information Theoretic Learning Methods for MDPs with Parametric Uncertainty"
    Advisor: Archis Ghate
  • B.Tech in Aerospace Engineering with Physics
    Indian Institute of Technology Madras
    2008 – 2013
    Thesis: "Hierarchical Decision Making using Spatio-Temporal Abstraction in RL"
    Advisor: Balaraman Ravindran

Press & Media

Subscribe

Get updates on new research, writing, and projects.