Sourabh Kumar

I'm a Data Scientist with a deep passion for building intelligent, scalable, and production-grade AI systems. With a strong foundation in machine learning, deep learning, and the rapidly evolving field of large language models (LLMs), I bring a versatile and solution-oriented approach to solving real-world problems. My technical skill set includes hands-on experience with modern frameworks such as LangChain, Hugging Face Transformers, PyTorch, TensorFlow, and robust deployment tools like Docker, Kubernetes, and Kubeflow.

Throughout my professional journey, I’ve had the opportunity to work across different domains, including document digitization, semantic search, multi-agent chat systems, MLOps pipelines, and time series forecasting. At Rsystems, I led the development of advanced AI models, including fine-tuning Llama3.2 and LayoutLMV3 using QLora with the Unsloth framework for tasks like NER tagging and document classification. I built a powerful multi-agent chatbot solution capable of handling structured, unstructured, and visual data, all deployed using container orchestration tools for scale and resilience.

My work in MLOps involved building an end-to-end pipeline using Kubeflow for experiment tracking, CI/CD, model deployment, and drift detection. This included managing versioned data with MinIO, real-time inference using Seldon Core, and implementing monitoring dashboards with Prometheus, Grafana, and Alibi Detect.

At Optum, I contributed to AI-driven automation in the healthcare and insurance domain. I designed a semantic search engine with ElasticSearch and transformer-based embeddings, built NER pipelines with finetuned LLMs, and architected solutions for contract segmentation using a combination of Vision Transformers and OCR. My projects also included pharmacy network digitization using YOLO, BERT, and LangChain pipelines, and forecasting claim volumes using a variety of time series models like ARIMA, SARIMA, and LSTM.

I strive to stay ahead of the curve through continuous learning, completing certifications in areas such as Generative AI, Azure Solutions Architecture, and MLOps. I take pride in bridging the gap between research and real-world applications, ensuring that every AI system I build is not only cutting-edge but also robust, efficient, and maintainable.

With a problem-solving mindset and a strong grip on end-to-end AI system design, I’m passionate about leveraging technology to drive meaningful impact across industries.