$ planopt

This is a project that helps business users: they chat with an AI agent to get forecasts and simulations of their sales and production quantities. The AI model is given MCP tools that run the forecasting and simulation, making it more accurate and robust — without needing an in-house data scientist to maintain it.

This is company IP, so there’s no public repo — but I’m happy to talk about the design in detail.

Flow

1. Forecast — LightGBM models are trained on historical sales data to predict demand per product.
2. Optimize — the forecasts feed a PuLP linear program that produces a production plan under real-world capacity constraints.
3. Interface — an MCP server exposes the pipeline to a chat interface, so a planner can ask “what should we produce next week?” instead of reading spreadsheets.

What I found fun

I find this project very fun to work on, since it’s quite a practical case of using AI to make things more approachable for users, while still retaining accuracy by using ML models and Python scripts.

In my opinion, this gives the user more accurate output to judge from, while being much easier to use since it’s a chatbot.