Our systems work is interdisciplinary, drawing across colleagues with expertise in programming languages, systems, networking, cryptography, privacy and security, and domain specialization. We break silos and innovate across various layers in the systems stack spanning hardware, networks, storage, and compute. We are particularly interested in improving and leveraging the foundational large language models (LLMs) to tackle the most interesting and challenging research problems in computer systems.
Our ongoing research projects include the following:
AI-Driven Software Engineering: Generative AI is transforming the way software is built. We conduct research at the forefront of this transformation. We design ML models, algorithms and platforms to improve developer productivity and software reliability.
AI Infrastructure: Systems optimizations are crucial for unlocking the true potential of AI-powered applications because the AI/ML workloads are extremely expensive to train and serve. The AI Infrastructure team at Microsoft Research India works on cutting-edge systems optimizations for improving the efficiency AI/ML workloads e.g., LLMs.
Network Brain: The Network Brain project focuses on holistic optimizations of large-scale networked services to improve their performance, reliability, efficiency and more. Our approach is based on a logically centralized “network brain”, which pools together information from myriad sources to make informed decisions.
EzPC: We work at the intersection of cryptography and systems, to design new performant secure multi-party computation protocols, with a specific focus on privacy preserving machine learning applications
HyWay: We believe that hybrid interaction – a key facet of post-covid workspaces – is more than being just about meetings. HyWay (short for Hybrid Hallway) enables mingling between in-person (physical) and remote (virtual) users in semi-structured and unstructured settings.
Domain Specialization: Specialized hardware such as GPUs and FPGAs have received increasing adoption with the end of Moore’s law, and this trend is likely to continue over the next decade. This project focuses of building domain-specific solutions for data systems — a class of systems where performance scaling is going to be of utmost importance as the data sizes grow.