Nirupama Balasubramanian

• Engineered an Autonomous AI Agent using Python, Java Spring Boot, and MCP to orchestrate LLM-driven graph reconstructions and shipment rate refreshes, targeting a 70% reduction in manual supply chain lookups

• Developed and fine-tuned Faster R-CNN and YOLOv8 object detection models for India's Smart City roadside monitoring initiative, reducing manual inspection workload by 40%
• Built a computer vision inference pipeline for batch and real-time workloads, deploying models using TorchServe for scalable serving

• Developed a medical CV pipeline using Vision Transformer and TypeScript; analysed real-time inference via Flask REST API to achieve 96% accuracy through research-led fine-tuning
• Architected a microservice to deploy an SVM-based predictive engine for a Tier-2 Singaporean Telecom provider; optimized high-dimensional feature sets to achieve 92% accuracy while leading a 45-member technical cohort

• Automated sentiment extraction from employee feedback, by designing and training a BERT-based transformer model with end-to-end NLP pipelines, reducing manual analysis effort and achieving 90% model accuracy

Deep Learning · Image Processing · Image Enhancement

• Authored a Diagnosis based Image Quality Assessment and enhancement technique for Low Dose CT Scans leveraging feature extraction and spatial enhancement methodology
• The SVR-GS model achieved a 0.969 PLCC, with enhancements verified by Dr. Inthulan Thiraviraj

PyTorch · XLM-R · NLLB-200

• Architected an XLM-RoBERTa framework across 9 languages using NLLB-200 for translation-based oversampling to resolve manifold collapse and increase Macro F1 from 0.63 to 0.727
• Integrated Supervised Contrastive Learning and Focal Loss to disentangle overlapping rhetorical labels, achieving 8% boost through Dynamic Thresholding

ML · Python · PySpark · TensorFlow

• Predicted Flight Delays for over 1 million records using Catalyst Optimizer and Lazy Evaluation
• Optimized to 0.6535 AUC via Grid Search and validated scalability by benchmarking PySpark against Pandas

Published in Advances in Electrical and Computer Technologies (CRC Press / Taylor & Francis)

Published in 2025 International Conference on Information, Implementation, and Innovation in Technology (I2ITCON) — IEEE

Foundational cloud architecture, security, and AWS service knowledge.

Design of scalable, secure, and cost-optimized cloud infrastructure.