Hardik Chauhan — Senior Applied Scientist & ML Engineer

About

I'm a Senior Applied Scientist at Microsoft Turing, where I lead the development of voice AI for Microsoft 365 Copilot. My work sits at the intersection of speech processing, large language models, and production systems.

MS in Computer Science from UW-Madison (full scholarship). B.Tech from IIT Roorkee. Previously ML Engineer at ExaWizards (Tokyo), research at IIT Patna and Naver (South Korea).

Published at AAAI, ACL, COLING, EMNLP. Reviewer for NeurIPS, ICML, ICLR, AISTATS, EMNLP, ACL.

Speech & Audio AI LLM Post-Training Diffusion Models NLP Multimodal Learning Evaluation & Safety PyTorch Python C/C++ Real-time APIs Production ML

Reviewer

NeurIPS '24 '25 ICML '25 ICLR '25 AISTATS '25 '26 ACL '24 EMNLP '23

Experience

Technical Lead, Voice Mode

Microsoft Turing · Redmond, WA

Productization & Launch

Led Microsoft 365 Copilot Voice Mode from early demo to production at million-user scale, defining model requirements and guiding real-time API integration with engineering.
Led Microsoft's first GPT-4o-based custom voice fine-tuning efforts, defining end-to-end data strategy and shipping production custom voice, featured in keynote demo at Microsoft Ignite 2025.

Evaluation & Safety

Designed and standardized an end-to-end voice quality stack (human evals, automated evals, production telemetry) that became the default evaluation path across Microsoft voice efforts.
Established production voice safety by building red-team datasets, designing audio-specific safety evals, and partnering with Responsible AI to define launch criteria.

Modeling & Post-training

Led development of diffusion-based audio decoders, improving speaker fidelity, prosody consistency, and production robustness.
Led audio post-training (SFT + RL), building datasets, reward signals, and evals to improve search triggering and various production capabilities.

Machine Learning Engineer

ExaWizards · Tokyo, Japan

Jan 2020 — Aug 2021

Created in-house dataset of 8.8M Japanese documents and trained ColBERT for document search.

Research Assistant

IIT Patna · India

Nov 2018 — Nov 2019

Designed neural architectures for multi-emotion controllable response generation. Published 5 papers at AAAI, ACL, COLING, IEEE TAC.

Research Intern

Naver Clova AI · Seoul

Summer 2018

NLP research with Dr. Jung Woo Ha and Minjoon Seo.

Selected Publications

EMNLP 2023

DUBLIN: Document Understanding By Language-Image Network

K Aggarwal, A Khandelwal, … Hardik Chauhan, et al.

Paper ↗

AAAI 2021

More the Merrier: Multi-Emotion and Intensity Controllable Response Generation

M Firdaus*, H Chauhan*, A Ekbal, P Bhattacharyya

Details ↗

COLING 2020 Oral

MEISD: Multimodal Multi-Label Emotion, Intensity and Sentiment Dialogue Dataset

M Firdaus*, H Chauhan*, A Ekbal, P Bhattacharyya

Details ↗

COLING 2020

Reinforced Multi-task Approach for Multi-hop Question Generation

D Gupta*, H Chauhan*, RT Akella, A Ekbal, P Bhattacharyya

Details ↗

ACL 2019

Ordinal and Attribute Aware Response Generation in a Multimodal Dialogue System

H Chauhan*, M Firdaus, A Ekbal, P Bhattacharyya

PDF ↗

IEEE TAC

EmoSen: Generating Sentiment and Emotion Controlled Responses in a Multimodal Dialogue System

M Firdaus*, H Chauhan*, A Ekbal, P Bhattacharyya

Details ↗

Education

MS in Computer Science

University of Wisconsin-Madison

2021 — 2022 · Full Scholarship

Advanced Deep LearningAdvanced NLPTheoretical MLBig Data SystemsParallel Computing

B.Tech in Electrical Engineering

Indian Institute of Technology Roorkee

2014 — 2018 · Co-Founded DL research group

Let's Connect

chauhanhardik23@gmail.com LinkedIn GitHub