← rajarshi-ray29.github.ioBack
Download PDF
Rajarshi Ray
Staff Engineer, Artificial Intelligence at IBM · Part-Time Entrepreneur, SceneSenseAI
Nationality: USA Stony Brook, New York +1 (631) 949-3136 rajarshi5675@gmail.com rajarshi-ray29.github.io github.com/rajarshi-ray29 linkedin/rajarshi-ray

Education

M.S., Computer Science · Stony Brook University (SUNY), New York, USA
Aug 2024 – Dec 2025
GPA 3.9/4.0 · Coursework: Computer Vision, NLP, AI, Data Science
B.E., Computer Science · Birla Institute of Technology and Science (BITS), Pilani, India
2020 – Jul 2024
GPA 8/10 · Teaching Assistant, Object-Oriented Programming

Research Experience

Staff Engineer, Artificial Intelligence · IBM / IBM Research, Poughkeepsie, NY
Feb 2026 – Present
  • InspectionBench (with IBM Research): perception-grounded benchmark for embodied industrial anomaly verification, evaluating 4 agent architectures across 4,000+ real-world asset-inspection scenarios; manuscript targeting ICLR 2027.
  • Patent-pending agentic AI system generating 100K+ persona-driven synthetic financial transactions per run for large-scale workload testing on IBM z17 LinuxONE, fully on-prem for enterprise security.
  • Operate a local AI research playground on IBM GPU infrastructure: 10 self-hosted LLMs serving 20–25 engineers.
Graduate Researcher · Stony Brook University, VAI-GM Lab, New York, USA
Aug 2024 – Feb 2026
Advised by Prof. Klaus Mueller
  • NTIRE 2025 Super-Resolution (×4) Challenge: led Team VAI-GM; designed MDRCT (Multi-step Dense Residual Connected Transformer) extending the DRCT baseline; achieved 30.86 dB PSNR / 0.8942 SSIM, ranked 18th globally.
  • 3D lung CT synthesis: generated high-fidelity synthetic CT volumes via VQ-VAE latent encoding and guided diffusion; deployed a Dockerized Orthanc DICOM server for secure image storage.
  • All-in-one image restoration: curated 80,000+ grid images (rain, haze, noise, snow) and trained a single transformer to restore multiple degradations.

Publications & Technical Writing

Ray, R., et al. "InspectionBench: Embodied Industrial Anomaly Verification with Perception-Grounded Physical AI Evaluation." In preparation (with IBM Research), targeting ICLR 2027.
Chen, Z., Ray, R., Tomar, S.S., Mueller, K., et al. (2025). "NTIRE 2025 Challenge on Image Super-Resolution (×4): Methods and Results." CVPR Workshops (NTIRE). arXiv:2504.14582
Ray, R. (2026). "From Prompt to Pipeline: AI-Powered Transaction Generation and Workload Testing on IBM Z LinuxONE." IBM Z & LinuxONE Community. IBM Community
Ray, R. (2026). "Synthetic Transaction Generation Powered by AI Agents." IBM Z & LinuxONE Community. IBM Community
Ray, R. (2025). "Porting and Deployment of the AWS Retail Store Application on IBM s390x using OpenShift." IBM Z & LinuxONE Community. IBM Community

Industry & Entrepreneurial Experience

Part-Time Entrepreneur · SceneSenseAI, New York, USA
2026 – Present
  • Building personalized AI teaching assistants over lecture video libraries: learners ask questions in natural language and get answers cited to exact timestamps in the source videos.
  • Designed an instructor-facing analytics engine surfacing learner confusion points and content gaps per module; validating with course creators and university instructors through a US pilot program.
Research and Development Intern · IUDX, Indian Institute of Science (IISc), Bangalore, India
Jan 2024 – Jul 2024
  • Developed a secure inference workflow for pneumonia classification on AMD SEV-SNP, a 20% performance gain over comparable stacks (Intel SGX, AWS Nitro) while preserving data privacy and code integrity.
  • Designed a confidential medical image analysis pipeline integrating TPM-based storage with remote inference (TensorFlow, Keras, C++, Docker).

Technical Skills

AI / ML: PyTorch, TensorFlow, Diffusion Models, VAEs, GANs, Transformers, LLMs, Agentic AI, RAG, Hugging Face, OpenCV, YOLO
Systems: Docker, Podman, OpenShift/Tekton, Kubernetes, Linux (x86/s390x), JMeter, Confidential Computing (SEV-SNP, TPM), Orthanc/DICOM
Languages: Python, C/C++, Java, Kotlin, SQL, Bash