An end-to-end MLOps pipeline for loan approval, built to mirror a standard "manual exploratory workflow -> automated pipeline" architecture:

Manual/Exploratory: Ingest -> Prepare/Transform -> Process -> Feature Store -> Bias Detection -> Train & Tune -> Register Model -> Deploy -> Online Feature Store -> Run Inference

Automated: Jenkins Pipeline -> Monitor (drift) -> Retrain -> Scale Inference

Everything runs locally with pip + venv. No Docker, no WSL, no cloud account required.

cd C:\
git clone <your-repo-or-just-copy-this-folder>   REM or just copy the folder to C:\ml-mlops-demo
cd ml-mlops-demo

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

Install Jenkins separately: download the Windows installer from https://www.jenkins.io/download/ , run it (it installs as a Windows service on port 8080), then open http://localhost:8080 and finish the setup wizard (it'll give you an initial admin password from a file path it shows you).

Open 5 separate PowerShell windows, cd C:\ml-mlops-demo and venv\Scripts\activate in each one, then:

Window 1 — MLflow tracking server (leave running)

mlflow server --host 127.0.0.1 --port 5000

Open http://localhost:5000 in a browser — you should see the MLflow UI.

Window 2 — run the manual/exploratory pipeline once

python src\ingest.py
python src\prepare_features.py
python src\feature_store\materialize.py
python src\train.py
python src\bias_check.py

Check MLflow UI: Experiments tab shows loan_approval with 3 runs (random_forest, gradient_boosting, bias_audit). Models tab shows loan_approval_model with a champion alias.

Window 2 — start the inference API (leave running)

python -m uvicorn src.deploy_api:app --host 127.0.0.1 --port 8001

Open http://localhost:8001/docs — try POST /predict with {"applicant_id": 5}.

Window 3 — generate a monitoring report

python src\monitor.py

Open reports\monitoring\drift_report.html in a browser — this is your Evidently drift dashboard.

Window 3 — trigger a full retrain (this is your "automated" story)

python src\retrain_trigger.py

Watch it re-ingest, re-train, re-check bias, and hot-swap the running API in Window 2 to the new model — call /health on port 8001 again and watch model_version increment with zero downtime.

Window 4 & 5 — scale inference demo

REM Window 4
python -m uvicorn src.deploy_api:app --host 127.0.0.1 --port 8002

REM Window 5
python src\scale_inference_demo.py

Shows requests round-robining across two replicas.

1. In Jenkins, New Item -> Pipeline, name it loan-approval-pipeline
2. Under Pipeline, choose "Pipeline script", paste the contents of jenkins\Jenkinsfile
3. Edit the PROJECT_DIR env var at the top of the Jenkinsfile to match your actual path (e.g. C:\ml-mlops-demo)
4. Click Build Now — watch each stage run in the Jenkins UI (Ingest -> Prepare -> Feature Store -> Train -> Bias Check -> Deploy -> Monitor)
5. For the "Retrain" story: duplicate the job, point it at a Jenkinsfile that just runs retrain_trigger.py, and set a Build Trigger -> Build periodically schedule (e.g. H */6 * * *) to show you understand scheduled retraining

Honest framing for the interviewer (use this if asked "is this production-grade?"): this is a local, single-node reference implementation of the pattern — in production you'd swap SQLite/file stores for Redis/S3/a real data warehouse, Jenkins for a managed orchestrator or Airflow-on-k8s, and add authentication, autoscaling, and canary deployments. The point was to demonstrate you understand every stage of the lifecycle and how they connect, not to reinvent SageMaker.

• feast, mlflow, and the API all show harmless DeprecationWarning lines on some versions — cosmetic, doesn't affect functionality
• The bias-mitigation step (ThresholdOptimizer) needs the sensitive attribute (gender) passed at prediction time too — that's realistic: fairness-aware post-processing generally requires it, and it's a good talking point about the practical trade-offs of different mitigation techniques (pre-processing vs. in-processing vs. post-processing)
• If port 5000/8001/8002/8080 are already in use on your machine, change the port in the relevant command AND in deploy_api.py / retrain_trigger.py (MLFLOW_TRACKING_URI, INFERENCE_API_URL)