Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/abdiasarsene/routerwise-server-predictive-analytics-for-shipments

RouterWise is a production-grade, modular MLOps pipeline built to optimize supply chains in the sensitive and prestigious art logistics sector. It is the intelligent backend of the upcoming PrecisioArt platform (Django-based), and delivers predictive insights and smart routing through a fully automated, monitored, and versioned ML lifecycle.
https://github.com/abdiasarsene/routerwise-server-predictive-analytics-for-shipments

bentoml catboost mlflow prometheus scikit-learn shap

Last synced: 2 months ago
JSON representation

RouterWise is a production-grade, modular MLOps pipeline built to optimize supply chains in the sensitive and prestigious art logistics sector. It is the intelligent backend of the upcoming PrecisioArt platform (Django-based), and delivers predictive insights and smart routing through a fully automated, monitored, and versioned ML lifecycle.

Awesome Lists containing this project

README 

          
# 🎯 RouterWise — ML-Powered Logistics Optimization for Art Handling



RouterWise is a production-grade, modular MLOps pipeline built to optimize supply chains in the **sensitive and prestigious art logistics sector**. It is the intelligent backend of the upcoming `PrecisioArt` platform (Django-based), and delivers predictive insights and smart routing through a fully automated, monitored, and versioned ML lifecycle.



---



## 🧠 Project Purpose



Designed for high-value and highly constrained logistics workflows, this pipeline ensures:

- Predictive analytics for routing and optimization

- Clean, auditable MLOps with Continuous Training (CT)

- Modular, production-first codebase for scalability



---

![API Predictiver](./statics/postman.png)

---



## 🧰 Tech Stack



| Area                  | Tools & Frameworks                                               |

|-----------------------|------------------------------------------------------------------|

| Data Versioning       | DVC                                                              |

| Feature Selection     | CatBoostEncoder, RobustScaler, VarianceThreshold, Chi², RFE     |

| Modeling              | Scikit-learn, CatBoost, Pipelines                                |

| Experiment Tracking   | MLflow (local & remote registry)                                |

| Model Packaging       | BentoML                                                          |

| API Serving           | FastAPI (RouterWise module) integrated with PrecisioArt (Django)|

| CI/CD                 | Jenkins, Makefile                                                |

| Monitoring            | Prometheus + Grafana                                             |

| Testing               | Pytest                                                           |

| Containerization      | Docker, docker-compose                                           |

| Retraining            | `retrain/` (upcoming), Celery + Beat                            |



![Feature Selection](./statics/feature_selection.png)

---

## 🏗️ Modular Project Architecture



```



routerwise/

│

├── app/                  # FastAPI app for serving predictions (BentoML runtime)

├── train\_pipeline/       # Feature engineering, training, inference, model saving

├── retrain/              # (WIP) Scheduled retraining logic with Celery + Beat

├── notebook/             # EDA and feature selection experiments

├── docker/               # Custom Dockerfiles

├── tests/                # Unit/integration test suites

├── Jenkinsfile           # CI/CD pipeline config

├── Makefile              # Unified entrypoint for all tasks

├── dataset.dvc           # DVC-tracked dataset pointer

└── README.md



````



---



## 🔁 MLOps Workflow



1. **Data versioning** using `DVC`

2. **Advanced feature engineering** with:

   - `CatBoostEncoder`, `RobustScaler`

   - `VarianceThreshold`, `Chi²`, `RandomForest`, `RFE`

3. **Training pipeline**:

   - Modular `sklearn` pipelines

   - Auto-logging to `MLflow`

4. **Model registration**:

   - Best model pushed to `MLflow Registry`

5. **Packaging and serving**:

   - Packaged using `BentoML`

   - Served via `FastAPI` (`app/`) — integrated with Django project `PrecisioArt`

6. **Monitoring & observability**:

   - Prometheus metrics collection

   - Grafana dashboards for API health, latency, drift, etc.

7. **Testing & validation**:

   - Unit and integration tests for both training and serving

8. **CI/CD**:

   - Automated with `Jenkinsfile`, lint/test/build/deploy



![Mlflow & BentoML](./statics/api.png)

---



## 🔄 Continuous Training Strategy (Planned)



A `retrain/` module is planned for scheduled model updates using **Celery + Beat**.  

Key points:

- New data triggers a scheduled pipeline

- Retrained model is **compared** to the currently deployed one

- **Only if the new model outperforms** the current one will it be promoted

- Else, the system retains the existing model



---



## 📊 Monitoring Capabilities



Deployed metrics collected in real time:

- API latency, health, uptime (Prometheus)

- Request counts, error rates

- Drift detection on incoming data streams

- Data quality checks on inputs



Visualized via **Grafana dashboards**. **(CLICK ON THE IMAGE BELOW TO WATCH THE VIDEO)**



[![Dashboard Preview](./statics/grafana_preview.png)](https://drive.google.com/file/d/1uD0oQKDrmADOqS0NHQR6PEfOGW2Jhqwu/view?usp=drive_link)



---



## ✅ CI/CD Pipeline



All components integrated into a production-grade `Jenkinsfile`:

- ✅ Unit tests

- ✅ Lint checks

- ✅ Build Docker image

- ✅ Trigger MLflow or BentoML packaging

- ✅ Optional deploy phase

- ✅ Slack/Webhook notifications (optional)



---



## ⚙️ Makefile Commands



```bash

make train       # Train and log with MLflow

make test        # Run test suite

make run         # Launch BentoML API server

make deploy      # Build + push containers

make monitoring  # Start Prometheus + Grafana stack

make format      # Run flake8 or ruff

````

![Makefile](./statics/makefile.png)

---



## 🔒 Reproducibility & Integrity



* Reproducible pipelines with versioned datasets (`DVC`)

* Isolated & dockerized environments

* Centralized logging via `MLflow`

* Strict test coverage

* Modular architecture for easy refactor or integration



---



## 📍 Status



* ✅ Training pipeline complete

* ✅ Feature selection logic optimized

* ✅ API (RouterWise) deployed via BentoML

* ✅ Monitoring dashboards active

* 🔜 Continuous Training module (`retrain/`) in progress



---



## 🤝 Contribution



This is not a starter project, but an evolving production-grade pipeline.

---



## 🔗 About



Built by **Abdias Arsène**, IT Consultant in AI & MLOps

Focused on real-world, cross-industry ML solutions (Health, Humanitarian, Finance, Art Logistics)



> *“I don't write code to run. I write code to endure.”*



---



![MLflow](./statics/mlflow.png)