Back
MLOps Architecture • ZenML • MLflow • FastAPI • Docker

Student Performance MLOps.

An end-to-end production MLOps engineering pipeline automating model training, experiment tracking via MLflow, and containerized inference deployment using FastAPI and Docker.

GitHub All Projects
Student Performance MLOps banner

Machine Learning Lifecycle Challenges

Many machine learning initiatives stall in exploratory notebooks, failing to deliver tangible business value due to the absence of automated production pipelines. Key bottlenecks include brittle experiment reproducibility, manual model versioning, dependency drift between local development and production servers, and the lack of automated inference services.

This project establishes an end-to-end production MLOps pipeline for academic student performance prediction, automating the complete lifecycle from data ingestion and feature engineering to experiment registry and containerized microservice deployment.

Automated Pipeline Architecture with ZenML & MLflow

The pipeline is orchestrated with ZenML, partitioning the workflow into five isolated, modular, and unit-testable steps:

[1. Data Ingestion Step] → Ingest raw datasets & validate schema integrity
  └── [2. Data Cleaning Step] → Missing value imputation, one-hot encoding, scaling
  └── [3. Model Training Step] → Train ensemble regression models with hyperparameter tuning
  └── [4. Model Evaluation Step] → Audit RMSE, MAE, R² score metrics & evaluate quality gate
  └── [5. Model Deployment Trigger] → Automatic promotion to MLflow Production Registry upon passing quality gate

Every pipeline run is automatically logged to a centralized MLflow tracking server, capturing hyperparameters, model artifacts, and evaluation metrics for complete auditability.

REST API Serving via FastAPI & Docker

Models passing automated quality gates are served through a high-performance FastAPI microservice:

POST/predict 
# Sample Request JSON Payload
{
  "gender": "female",
  "race_ethnicity": "group B",
  "parental_level_of_education": "bachelor's degree",
  "lunch": "standard",
  "test_preparation_course": "completed",
  "reading_score": 85,
  "writing_score": 88
}

# Model Output Response
{
  "predicted_math_score": 83.42,
  "model_version": "v1.0.4-prod",
  "latency_ms": 11.2
}

The service is containerized into a lightweight Docker image ready for deployment on cloud orchestrators (e.g., Kubernetes or AWS ECS) with dedicated health check endpoints and latency telemetry.

Back to Projects Back to top ↑