Nikhil Rathaur

I’m a Machine Learning, Deep Learning, and MLOps enthusiast with a foundation in Electronics & Communication Engineering and a Specialization in Rail Engineering. I’m motivated by building scalable, production-minded AI systems that turn data into intelligent automation for real engineering constraints.

Through internships with National High Speed Rail Corporation Limited (NHSRCL) and Banaras Locomotive Works (BLW), I gained exposure to railway signaling systems, infrastructure design, project lifecycle, safety systems, communication technologies, and maintenance workflows with fault detection. That domain perspective drives my focus on reliable, end-to-end solutions rather than isolated models.

In my projects, I build complete pipelines: for “Movie Sentiment Analysis using MLOps,” I delivered an end-to-end NLP system with modular preprocessing and TF-IDF features, experiment tracking and reproducibility via MLflow/DVC/DagsHub, and CI/CD using GitHub Actions. I containerized the Flask app with Docker, deployed it on AWS EKS using Kubernetes, and added Prometheus & Grafana for monitoring and observability—achieving 92% accuracy.

I also enjoy deploying AI to constrained and multimodal environments. With “Plant Disease Detection using TinyML & ESP32-CAM,” I trained and quantized a lightweight CNN into an 18 KB INT8 TensorFlow Lite model and implemented an OpenCV + Streamlit + ESP32-CAM pipeline using Edge Impulse and Arduino IDE for offline, low-latency inference (≈94% validation accuracy). Alongside that, I developed a FastAPI-based REST API with Pydantic for fast, validated CRUD operations (sub-200ms), and a multimodal Video Question Answering system using Whisper, semantic retrieval, and a Streamlit UI (93% retrieval accuracy).