Projects
BeanClassifier KF
• Built an end-to-end ML training and serving workflow using the Beans dataset (three-class leaf disease classification) and automated the entire process with Kubeflow Pipelines.
• Designed modular Kubeflow components (data preprocessing , training ,evaluation , model exporting) and containerized each stage using Docker for fully reproducible pipeline runs.
• Logged model metrics and artifacts to MLflow and stored final model weights in a GCS bucket for versioned, production-friendly model management.
• Developed a FastAPI inference service to serve the trained image classifier in real time, exposing a predict endpoint that processes uploaded leaf images and returns disease class probabilities.
• Deployed the serving API on Kubernetes with liveness/readiness probes and autoscaling enabled using HPA for reliable model inference under varying loads.
Finch – News application
Architecture
• Built an end-to-end CI/CD pipeline from source code to production deployment, implementing Docker multi-stage builds, image registry workflows, unit testing, Kubernetes deployments, and automatic scaling using HPA.
• Successfully predicted potential production failures and proactively resolved issues using monitoring insights, improving system stability and reliability.
github
QA Rag Pipeline
A Simple Rag Pipeline Using Langchain and chromavectordb , Ollama ,HuggingFace Embedding Models github
Architecture
