This project predicts whether a hotel reservation will be cancelled based on booking details. It demonstrates a full-stack MLOps workflow using scalable, modular components across development, experimentation, and production.

🔍 Built for reproducibility, automation, and real-time inference using best-in-class tools.

Tech Used: Pandas, GCS, Python

• Loads raw CSV data either from local or GCS buckets.
• Handles missing values, schema validation, and basic sanity checks.

Tech Used: Pandas, NumPy, Scikit-learn

• Transforms categorical data, encodes features, and scales numericals.
• Derives domain-specific metrics like total_nights, is_weekend, etc.

Tech Used: Scikit-learn, LightGBM, GridSearchCV

• Models: LightGBM, Random Forest, optionally XGBoost.
• Uses GridSearchCV for hyperparameter tuning.

Tech Used: MLflow

• Tracks runs, models, metrics, parameters, and artifacts.
• Registers best models for later deployment.

Tech Used: Docker

• Creates reproducible containers of the entire pipeline (training + inference).
• Ensures consistency across all environments.

Tech Used: Jenkins, Git, Docker, Kubernetes, Artifact Registry

• Jenkins triggers builds, trains models, creates Docker images, and deploys them to Kubernetes.
• Artifacts are stored in GCP’s Artifact Registry.

Tech Used: Google Kubernetes Engine (GKE), FastAPI, Streamlit

• Deploys trained model as an API (via FastAPI) or UI (via Streamlit).
• Handles real-time prediction requests.

Tech Planned: Prometheus, Grafana, Pub/Sub, Dataflow

• Monitor model inference latency, API uptime, and feature drift.
• Setup batch/streaming prediction workflows with Pub/Sub & Dataflow.

• Packages code and dependencies into isolated containers.
• Ensures reproducibility across dev, CI, and cloud.
• Integrates with Jenkins and Kubernetes effortlessly.

• Automates ML lifecycle from build → train → package → deploy.
• Reduces manual effort and allows scalable team collaboration.
• Pushes images to Artifact Registry and deploys on GKE.

• GCS (Cloud Storage): Stores datasets and trained models.
• Artifact Registry: Maintains Docker images used in production.
• GKE: Deploys and auto-scales the inference API.

• Logs and compares model training runs.
• Supports model registry for deployment.
• Improves transparency and reproducibility in training.

• Streamlit: Rapidly creates data-driven UIs for model demonstration.
• FastAPI: Provides fast, async REST APIs for serving predictions.

• 📊 Monitoring Dashboards: Use Prometheus + Grafana for performance monitoring.
• 📦 Kubeflow Pipelines: For orchestrated model training & deployment.
• 🔁 Batch/Streaming Inference: Using Pub/Sub + Dataflow pipelines.
• 📊 Admin Dashboard: Build a React + FastAPI admin panel for analytics & insights.

• ✅ CI/CD Workflow – Visualizing Jenkins → Docker → GKE flow
• ✅ Kubernetes Deployment Architecture – Pods, Services, Load Balancer
• ✅ ML Pipeline Flowchart – From ingestion to deployment
• ✅ Monitoring Stack Diagram – Prometheus + Grafana + AlertManager setup

Tilak Varma
AI Product Developer

📧 [varmatilak22@example.com]
🔗 LinkedIn • GitHub