Distributed AI systems engineer
I contribute upstream to the PyTorch ecosystem across collective communication, multi-node training, framework correctness, and runtime behavior. Compiler and runtime integration, ExecuTorch, and applied machine-learning research extend that work from GPU clusters to deployed systems.
01 / Systems
Distributed systems are the center of gravity. The surrounding work—runtimes, edge execution, diagnostics, and research—builds a broader understanding of what it takes to move models from development to dependable operation.
Primary focus
I work on the behavior that determines whether multi-node training is trustworthy: collective operations, transport layers, process-group lifecycle, backend coordination, and failure handling across heterogeneous hardware topologies.
The standard is simple: failures should be bounded and explainable, and successful work should preserve the correctness guarantees users expect.
I investigate correctness at the boundaries between autograd, tensor abstractions, checkpointing, parallel execution, and compiler-adjacent paths. The outcome is durable upstream behavior rather than local workarounds.
I contribute to portable operator behavior, kernel correctness, layout handling, and selective-build plumbing so inference can retain predictable semantics on resource-constrained targets.
I build tracing and integration paths that expose the route from framework APIs to underlying execution. Clear errors, precise tests, and useful profiling are part of the engineering surface.
02 / Applied ML
My research applies computer vision and machine learning to accessibility, sensing, resilience, and deployable intelligent systems.
2026
A Gloss-driven Indian Sign Language Production System Using Learned Pose Representations.
International Joint Conference on Artificial Intelligence (IJCAI-ECAI)
2026
MSBAD at SignEval 2026: Multi-Scale Born-Again Distillation Ensemble for Radar-Based Sign Language Recognition.
IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops
2026
RAVEN: A Rapid Agentic Vision Framework for Emergency Response in Vulnerable Settlements.
IEEE/CVF Winter Conference on Applications of Computer Vision Workshops
2026
Enhancing Earthquake Preparedness in the Himalayan Region: A Machine Learning Approach Using EEW System Parameters.
Iranian Journal of Science and Technology, Transactions of Electrical Engineering
2025
Hierarchical Windowed Graph Attention Transformer Encoder and a Large Scale Dataset for Indian Sign Language Recognition.
Pattern Analysis and Applications
2025
Decorrelation-Based Self-Supervised Visual Representation Learning for Writer Identification.
ACM Transactions on Asian and Low-Resource Language Information Processing
2025
A Continuous Indian Sign Language Production System from Isolated Signs.
International Conference on Smart Computing and Communications
2025
Edge-Based Intelligent and Smart Health Monitoring on PYNQ-Z2 Using Lightweight Protocol and Integration of Machine Learning Models.
International Journal on Smart Sensing and Intelligent Systems
03 / Upcoming
Accepted presentations and other forthcoming technical activity. This is kept deliberately small so meaningful updates remain easy to find.
October 2026
Accepted poster
Portable Diagnostics for LLM Inference on vLLM and llm-d
October 2026
Accepted poster
Accelerating Modern Attention Kernels with Triton-TLE
04 / Community
I serve as a reviewer for IEEE Transactions on Artificial Intelligence, the PyTorch blog, and the PyTorch Associate certification course. I have also contributed to program and CFP review for MCP developer events.
I speak about distributed communication, PyTorch internals, collective operations, and on-device inference at engineering meetups and universities, including BangPypers, Razorpay Engineering Meetup, the Red Hat PyTorch × Hugging Face Meetup, IISc, IIT Bombay, and import Bengaluru. I have also presented sign-language research at IJCAI-ECAI and mentored at the Hugging Face–Meta Hackathon.
I write technical material for Red Hat Developer on the distributed communication infrastructure evolving within PyTorch.
Contact
arkadipmaitra at gmail dot com