rajendra behera

I lead the design and delivery of production machine learning and generative AI systems that support the organization's mission to help people in need. My work spans the full lifecycle, from framing the problem and preparing data to shipping models and keeping them healthy in production. I build LLM powered applications using retrieval augmented generation, including embedding pipelines, vector search, reranking, and evaluation, so users get answers that are accurate and grounded. I also design the data pipelines, APIs, and automation that connect these systems to real workflows. Beyond the core engineering, I focus on reliability, observability, and cost efficiency so the platform scales responsibly. I work closely with product and operations to turn ambiguous requirements into working software, and I mentor teammates on modern AI tooling and practical best practices.

I designed end to end MLOps pipelines on GCP and Azure using Python, SQL, Docker, and GitHub Actions with PyTorch, TensorFlow, JAX, XGBoost, and Weights and Biases, cutting model deployment cycles by 37 percent and provisioning infrastructure through Terraform. I raised cybersecurity threat detection precision from 82 to 93 percent by fine tuning LLMs with NLP, data augmentation, and optimized embeddings, and I built RAG and prompt engineering workflows to strengthen model reliability across multi source intelligence data. I reduced fraud detection false positives by 64 percent, scaled compliance automation to cover 95 percent of CMMC requirements, and led a four person team building a pricing compliance system using reinforcement learning and policy modeling. I also designed A/B testing and causal analysis frameworks and delivered Power BI dashboards that improved reporting efficiency by 37 percent.

I deployed large scale classification models and data pipelines for atmospheric data analysis on distributed PyTorch and TensorFlow clusters running Hadoop and GPU, increasing prediction throughput by 64 percent to support climate modeling. I built sentiment analysis and CTR prediction pipelines using data mining on Hadoop and GCP, and I validated AI models across more than ten aircraft subsystems in close collaboration with cross disciplinary teams. I analyzed multi sensor telemetry from over 150 aircraft to surface climate related degradation trends, reducing unplanned downtime by 19 percent annually. I also optimized and deployed ONNX neural networks on embedded flight systems using computer vision, achieving a 28 percent throughput improvement at under two percent error. Alongside the engineering, I mentored three ML engineers on deployment, version control, and cloud scaling, which accelerated project delivery.