CV
Education
- B.Tech in Computer Science Engineering, Indian Institute of Technology Bhilai, 2019-2023 (GPA-9.66/10.00)
Work experience
- Apr 2026 - Present: Research SDE at Microsoft M365 Research
- Working on efficient LLM inference
- Jul 2024 - Apr 2026: Research Fellow at Microsoft Research India
- Designed an efficient & practical sparse attention technique Kascade.
- Impact: Kascade achieves up to 4.1x speedup for decode attention and 2.2x for prefill attention over FlashAttention-3 on H100 GPUs using custom kernels.
- Impact: Verified using large models like Qwen-3-32B, maintaining dense-model accuracy on reasoning benchmarks (AIME-24).
- Architected a modular evaluation framework using the Strategy Pattern to de couple attention implementations from models, enabling rapid benchmarking of sparse attention schemes via Transformers on a range of downstream tasks.
- Supervised finetuning using FSDP on MultiHop Question-Answering datasets to learn document-local sparse attention.
- Oct 2023 - Jun 2024: Associate Software Development Engineer at Publicis Sapient
- Modernized the Deposits journey on web and mobile for Lloyds Banking Group
- Spearheaded WCAG 2.1 accessibility for the application
- Feb 2023 - May 2023: UG Research Fellow at TIH IIT Bhilai
- Created the prototype for a Bike Renting System using LoRa technology
- LoRa Protocol for communication between Gateway and bike-locks
- Coding the App-Server system for renting the bike
- May 2022 - July 2022: Sofware Development Engineering Intern at Publicis Sapient
- Integrated Google PubSub communication in Point of Sale ordering and management system making it data flow more reliable
- Designed and Deployed end-to-end face recognition microservice 25% more efficient and 99% smaller network data packet size than earlier design
Achievements
- Award for Learning Mindset at client townhall while working at Publicis Sapient
- Smart India Hackathon’22 Finalist & Runner Up in Software Monitoring category
- Member of a team of 4 that stood 1892/9000+ teams in Google Hashcode’21
- Led a team that stood 279/3500+ teams in the ICPC’21 Amritapuri preliminary round and qualified for the regional round
Skills
- Languages: Python, C++, C, Javascript, CUDA, Go
- ML Technologies: Deep Graph Library(DGL), PyTorch, OpenCV, HuggingFace
- Tools: Git, Github, Docker(basic), Python Notebooks, AWS, GPGPUSim, Kubernetes, Microk8s, GCP
- Coursework: Data Structures, Algorithms, Machine Learning, Parallel Programming, Artificial Intelligence, Information Retrieval, Computer System Design, Natural Language Processing, Approximation Algorithms, Digital Image Processing, Distributed Systems
Projects
Distributed Graph Neural Networks
D. Deshmukh, G. R. Gupta, M. Chawla, V. Jatala and A. Haldar, "Entropy Aware Training for Fast and Accurate Distributed GNN," 2023 IEEE International Conference on Data Mining (ICDM), Shanghai, China, 2023, pp. 986-991
Kascade: A Practical Sparse Attention Method for Long-Context LLM Inference
D. Deshmukh, S. Goyal, N. Kwatra, and R. Ramjee, "Kascade: A Practical Sparse Attention Method for Long-Context LLM Inference," arXiv preprint arXiv:2512.16391, 2025
Teaching
Service and leadership
- May 2022-Mar 2023IIT Bhilai - DSAI Club Department Head computer Vision
- Spearheaded in projects related to computer vision
- Conducted workshops to introduce students to computer vision
- May 2021-Apr 2022 IIT, Bhilai - Centre for Career Planning Services Computer Science Representative
- Led a team of 4 to create and conduct 6 programming/technical tests to make final year students interview ready
- May 2021-Apr 2022 IIT, Bhilai - UG EECS Representative
- Guided students regarding academic decisions
- Faciliated smooth function of academic activites at insitute