Headline

AI Software Engineer @ Intel || Focus on GenAI and NLP || LLMs || Deep Learning || Machine Learning

About

I am an AI Software Engineer with focus on high-performance AI infrastructure and production-grade AI applications. At Intel, I specialize in the internals of PyTorch, optimizing distributed training (FSDP, DeepSpeed, MegatronLM) and inference runtimes (vLLM, SGLang) for custom AI accelerators. I have a proven track record of upstreaming complex features to the PyTorch ecosystem, such as Context Parallelism in TorchTune. I regularly work across the deep learning stack, including: • PyTorch internals and runtime systems (torch.compile, TorchDynamo, TorchInductor, autograd) • Distributed training and parallelism (FSDP, tensor, pipeline, expert, and context parallelism) • LLM inference systems and serving frameworks such as vLLM and SGLang • Accelerator-aware optimization for large-scale model training and inference • Quantization, KV-cache optimization, and long-context model execution • Building production AI applications, including LLM-powered APIs, RAG systems, semantic search, and scalable model-serving pipelines Beyond infrastructure, I have extensive experience architecting and deploying end-to-end AI applications. I have built and scaled NLP systems—including semantic search engines and automated insurance claim processors—that serve tens of thousands of users and deliver significant operational efficiency. These systems involved model development, deployment pipelines, monitoring, and scalable APIs. I enjoy working on problems where systems engineering meets applied AI. I have successfully collaborated across teams to deliver high-quality AI solutions within tight deadlines, always focusing on continuous improvement and learning. As a quick learner and proactive problem-solver, I am excited to take on new challenges and contribute to innovative AI and machine learning projects. GitHub: github.com/SKRohit Blog: skrohit.github.io

Location

Bengaluru

AI stack

claude code

claude

github copilot

hugging face

kimi

vLLM

Work experience

AI Software Engineer at Intel Corporation, Jun 2022 - Present · 4 yrs 4 mos, Bengaluru, Karnataka, India, As part of the Intel AI Frameworks team, I work on distributed AI systems and open-source LLM infrastructure, focusing on high‑performance training and inference on Intel accelerators. I optimized large‑scale Llama 3 pre‑training and fine‑tuning with DeepSpeed and PyTorch, and deployed low‑latency, high‑throughput inference using SGLang and vLLM on AWS. I also contributed to core framework features by adding context‑parallel support to TorchTune and FP8 support to SGLang for Intel Gaudi, and built performance projection tooling using PyTorch execution traces, Docker, Kubernetes, MLflow, and W&B to evaluate LLM workloads on Intel hardware.

NLP Software Engineer at Carelon Global Solutions India, Nov 2020 - May 2022 · 1 yr 7 mos, Bangalore Urban, Karnataka, India, As a part of the Cognitive AI team, I was responsible for improving in‑house NLP and machine learning models, and moving them into production. Apart from supervised learning I worked on some semi-supervised learning approaches e.g. SimCSE, SCAN loss, etc as well. Also, I created a semantic search Python module for efficiently identifying relevant medical keywords from a large medical corpus.

Machine Learning Engineer, NLP at Zycus, May 2020 - Nov 2020 · 7 mos, Bengaluru, Karnataka, India, Being part of the Product Development Team (PDT-1) at Zycus, I worked on sample efficient information extraction models to extract relevant information from news articles and legal documents that have been deployed and used by Zycus's clients. Also, I architected and deployed a multi-node Elasticsearch cluster to visualize the massive amount of text data using Kibana.

Systems Engineer at Infosys, Dec 2018 - May 2020 · 1 yr 6 mos, Bengaluru, Karnataka, India, As a Systems Engineer, I was part of the iCETS unit (R&D division of Infosys) where my primary role was to design POCs for computer vision and NLP tasks using the latest state of the art research in these domains. I worked majorly on the latest deep neural architectures and models like seq2seq networks, transformers, and GPT2, BERT, YOLOv3, RCNN networks respectively. During my tenure, I developed end to end neural object detection and text summarization POCs which are being used by Infosys and its clients. I was also awarded the Best Debutant Award for the year 2019 in iCETS unit.

Tutor at Concept Kendra, Feb 2018 - Dec 2018 · 11 mos, Hyderabad, Telangana, India, Worked as a Physics and Mathematics tutor teaching high school students.

Education

Bachelor’s Degree, National Institute of Technology Rourkela

High School, kendriya vidyalaya

Public links

github: https://github.com/SKRohit