I’m an applied AI researcher/engineer, currently at AI71 in Abu Dhabi. My work here centers on LLM agent systems for enterprise: agent orchestration, making them production-ready, and evaluation frameworks. Recently I’ve also started working on voice-based use cases.
Before this, I spent nearly four years at Wadhwani AI in Delhi, building AI for public health. The work I’m proudest of: a chatbot for frontline healthcare workers, automated TB diagnostics (AAAI 2024), and a clinical decision support system used in over 122 million patient consultations. I studied at IIIT Delhi, an interdisciplinary program across Computer Science, Economics, and Sociology.
Born and raised in Delhi, now learning to live on my own in Abu Dhabi. Outside work, I follow cricket, climb mountains, travel whenever I can, and chase good food.
Experience
Where I've worked, in reverse chronological order.
-
Senior ML Engineer Mar 2025 to present
AI71
Building AI agents for enterprise. Architected the transition from static RAG to ReAct-based LLM agents using LangGraph. Designed evaluation frameworks, optimized inference costs by 3x, and benchmarked frontier models for reasoning.
-
Associate ML Scientist II Aug 2021 to Mar 2025
Wadhwani AI
Built AI systems for public health at scale, including an LLM chatbot for frontline healthcare workers, automated TB diagnostics (AAAI 2024), and a clinical decision support system serving 122M+ consultations. Partnered with the Indian government, USAID, and the Gates Foundation.
-
Teaching Assistant, Econometrics Jan to May 2021
IIIT Delhi
Assisted Dr Gaurav Arora with the Econometrics II course. Created data assignments in R and evaluated quizzes and exams.
-
Research Intern 2020
Wadhwani AI
Helped build AI and data science solutions to predict non-adherence in TB patients. Wrote about the internship experience.
-
Research Intern 2019
Trivedi Centre for Political Data
Helped create datasets on legislators and election-related judgments.
-
B.Tech in CS, Economics & Sociology 2017 to 2021
IIIT Delhi
Pursued an interdisciplinary program at the intersection of Computer Science, Economics, and Sociology. Where I first got interested in applying ML to real-world problems.
Projects
Selected things I've built.
-
ReAct-based LLM agent system at AI71 with structured reasoning, dynamic tool orchestration, prompt and KV caching (18-35% TTFT reduction), and a full evaluation framework with MLflow tracking and CI/CD integration.
-
LLM-based assistant helping frontline health workers in India with clinical queries. 83% acceptance rate from domain experts. Reduced per-query cost 5-6x with synthetic data and caching. Built with the Ministry of Health, supported by USAID and the Gates Foundation.
-
Object detection system using DETR to read Line Probe Assay strips for Drug-Resistant TB, achieving >96% band accuracy. Designed a ground truth study with 9 expert microbiologists. Paper at AAAI 2024.
-
ML system for India's national telemedicine platform, helping doctors identify the right diagnosis across 31 diseases. Served 122M+ consultations. Semi-finalist at MIT Solve Challenge 2023.
-
Corneal Transplant Outcome Prediction
AI suite predicting future visual scores and patient drop-off to improve cornea management and surgery outcomes. Led the project end-to-end, including product, stakeholder management, fundraising, and ML.
Publications
Beyond the desk
A few things I've done outside of my core work.
-
Speaking & Training 2021 to present
Talks, Panels & Workshops
Spoken at ACM COMPASS 2024 on AI for healthcare delivery in India, been part of panel discussions, and facilitated 8 AI training workshops for medical professionals, senior government officials, and partner agencies.
-
Field Visits 2021 to 2025
Travelling Across India for Fieldwork
While at Wadhwani AI, travelled to remote areas across multiple Indian states to talk to users, patients, doctors, and frontline health workers, collecting data, understanding ground realities, and shaping the AI products around real needs. One of the experiences I cherish the most.
Contact
Email is the best way to reach me at muk137kumar@gmail.com, though LinkedIn works too. I read everything and reply to most.