Data Scientist
Microsoft
Jun 2024 - Present
Pranjal Gulati
I'm a Data Scientist at Microsoft, working at the intersection of applied AI, machine learning, and large-scale data systems.
I'm interested in personalization, forecasting, trustworthy machine learning, and building systems that turn models into useful products.
Experience
Research Intern
Nyun AI
Dec 2023 - May 2024
Data & Applied Scientist Intern
Microsoft
May 2023 - Jul 2023
Research Intern
Video Analytics Lab, IISc
Aug 2022 - Oct 2022
Research Intern
AIShield, Bosch Global Software Technologies
May 2022 - Jun 2022
Research
My research spans federated learning, knowledge distillation, and trustworthy machine learning.
Clustered Federated Learning Framework for Non-IID Ambient Noise Mapping for Urban Sensing
International Joint Conference on Neural Networks (IJCNN), 2025
A clustered federated learning framework for learning urban noise maps from heterogeneous, distributed sensing data.
[Re] Distilling Knowledge via Knowledge Review
ML Reproducibility Challenge, 2021; arXiv preprint, 2022
A reproduction and empirical study of Knowledge Review, including ablations that examine how a student network learns from multiple teacher layers.
Open Source
I build and contribute to machine learning tools, experiments, and implementations. The current collection lives on GitHub.
Awards
Second place · Adobe Behavior Simulation Challenge
Dec 2023
Simulated social-media behavior and content using embedding-based and large language model approaches.
Second place · Microsoft Membership Inference Competition at IEEE SaTML 2023
Feb 2023
Developed membership inference attacks for image and tabular datasets from first principles.
Third place · DevRev Domain-Specific Question Answering Challenge
Feb 2023
Built an efficient retrieval and reader pipeline using BM25, SBERT, DeBERTa, domain adaptation, quantization, and distillation.
First place · Embedded Security Challenge, CSAW 2022
Nov 2022
Solved AI security challenges spanning trojan insertion, watermarking, membership inference, and poisoning.
First place · Bosch Model Extraction Challenge for Video-Based Models
Mar 2022
Extracted Video Swin Transformer and MoViNet models on video action-recognition tasks.
Education
Indian Institute of Technology Roorkee
Bachelor of Technology in Electronics and Communication Engineering
2020 - 2024
Professional Service
Reviewer, IJCNN 2027
Main Track · upcoming