Hire Om G. - Fine-Tuning Engineer

Own the full AI/ML lifecycle of a production image classification system for product identification — E2E responsibility from research and architecture through deployment on Azure-native infrastructure (Azure Container App Jobs, Azure ML, Azure AI Foundry). - Building a 3D-conditioned synthetic image generation pipeline and running large-scale experiments on Azure batch infrastructure. - Conducting applied research in generative AI forensics: analyzing distributional differences between synthetic and real image embeddings, grounded in spectral artifact and frequency-domain literature. - Architected an Azure OpenAI–based text classification service with prompt engineering, structured JSON outputs, Git-managed prompt versioning, and production monitoring via Azure AI Foundry. - Designing a unified evaluation framework for AI agent systems, establishing standardized benchmarking and testing protocols across agentic workflows. - Building developer tooling including automated experiment logging, context-injection patterns, and AI-assisted coding workflows for team productivity.

* Designed and deployed LSTM-based predictive models using PyTorch for forecasting equipment failures from multi-sensor time-series data, * improving predictive maintenance accuracy by 30% across IoT infrastructure. * Implemented synthetic data augmentation using GANs to overcome extreme class imbalance in failure datasets, increasing anomaly detection * sensitivity and generalization. * Built and optimized real-time data pipelines using Apache Kafka and Apache Spark to process 10M+ sensor records daily (~2TB/month), reducing * detection latency by 18%. * Containerized ML models with Docker and orchestrated deployment on Azure Kubernetes Service (AKS), ensuring scalable and reliable inference * in hybrid cloud environments. * Automated ML retraining workflows via Jenkins CI/CD pipelines, integrating concept drift monitoring, model validation checkpoints, and rollback * mechanisms for production stability. * Developed multi-stage anomaly detection systems using Spark MLlib and ensemble techniques, enabling proactive fault prediction and alerting in * live sensor networks. * Enhanced customer support systems by training Transformer-based sentiment models on GCP, streamlining ticket triage, and improving * prioritization efficiency by 22%.