Balanced Pipeline Simulation

The Requirement:

If Mack Trucks expects X trucks to be sold in the next 3 years, then considering the trucks already in the production pipeline,

• How does the future pipeline look like?
• How many trucks are to be built every month?
• What should be the schedule of all these trucks, in order to meet this retail target, considering the production capacity, history, factors like holidays, factory shut downs and expected worker strikes.

How did we solve it?

• Time-series forecasting for every truck and order type, using statistical and ML methods Prophet (via Silverkite) and Kernel Density Estimation
• Queing theory for simulation
• A frontend for the Stakeholders to input in various scenarios and analyze then through analytical graphs and a simulation table
• Developed an allocation algorithm to balance out production across expected product timelines
• Built Power BI report for the stakeholders
• Deployed into internal services using docker and Azure Devops pipelines

What did we achieve?

• Reduced chances for holding loss due to excess inventory
• A plan ready to face the future demand
• A balance between the production and demand
• A 40 times faster and 15% more accurate estimation than the existing manual approach
• Suggestions of what needs to be done, given the shortcomings of the present conditions

Technologies:

• Azure Services: Data Factory, Databricks, pipelines, docker
• Programming Languages: Python
• Data Manipulation: Spark, pandas
• ML Models: Scikit-learn, Scipy
• MLOps: MLFlow
• Frontend: Streamlit
• Forecasting: Silverkite
• Reporting: PowerBI

Team:

• Abhimanyu Bellam and Kelechi Ikegwu (Manager)