// Research Work

Research & Publications

Exploring the intersection of privacy, deep learning, and real-world AI deployment.

Privacy-Preserving AI
Computer Vision
Generative AI & LLMs
MLOps Systems
// Papers & Work

Research Work

Privacy AI LLMs 2025

Privacy-Aware RAG Systems: Automated PII Detection and Masking in Enterprise Document Intelligence

Rajveer Singh — Shri Balwant Institute of Technology & Management (SBNITM) in collaboration with LPU AI/ML Lab

This paper presents a framework for building GDPR and HIPAA-compliant Retrieval-Augmented Generation (RAG) pipelines. We introduce a pre-ingestion PII scrubbing layer using NER-based entity detection that prevents sensitive information leakage through LLM responses, while maintaining retrieval accuracy within 3% of unmasked baselines.

Keywords: RAGPII DetectionGDPRLLM SafetyNER
⟳ Under Review / Preprint Pending (Expected Q3 2026)
Computer Vision Deep Learning 2024

Benchmarking CNN Architectures for Real-Time Facial Micro-Expression Recognition Across Multiple Datasets

Rajveer Singh — Shri Balwant Institute of Technology & Management (SBNITM) in collaboration with LPU AI/ML Lab

Comparative analysis of VGG16, ResNet50, EfficientNet, and MobileNetV3 for micro-expression classification on FER-2013, RAF-DB, and CK+ datasets. Transfer learning with EfficientNet-B3 achieved 87% accuracy on FER-2013, outperforming the ResNet baseline by 12 percentage points at 24 FPS inference speed.

Keywords: CNNTransfer LearningFEROpenCV
⟳ Under Review / Preprint Pending (Expected Q4 2026)
MLOps Production AI 2025

DLOps: A Modular Framework for End-to-End Deep Learning Pipeline Automation and Deployment

Rajveer Singh — Shri Balwant Institute of Technology & Management (SBNITM) in collaboration with LPU CSE Dept

Presents DLOps, a modular Python framework combining MLflow experiment tracking, Docker containerization, and FastAPI serving to automate the full deep learning lifecycle. Evaluated on three production use cases showing measurable reduction in deployment overhead and improved reproducibility.

Keywords: MLOpsMLflowDockerFastAPI
⟳ Under Review / Preprint Pending (Expected Q4 2026)