Quick2Plan

Inspiration

Production planning in factories is still too manual, reactive, and hard to explain. We wanted to help planners turn complex demand, line constraints, and changeovers into fast decisions they can trust.

What it does

Quick2Plan finds optimized weekly production sequences across canning lines. It reduces costly changeovers, compares real vs optimized plans, simulates OEE impact, handles what-if disruptions, and explains decisions through an AI chatbot.

How we built it

We model production as a graph: SKUs are nodes and changeovers are weighted edges. We use graph optimization for sequencing, CatBoost for production-time prediction, K-means for SKU pattern discovery, a deterministic simulator for fair OEE comparison, and a chatbot for explainability.

Challenges we ran into

A huge part of the project was understanding the real challenge behind the data. We spent a long time cleaning messy operational files, joining different sources, and figuring out what could be trusted. Another major challenge was designing an architecture that was not only powerful, but also explainable enough for planners and judges to understand.

Accomplishments that we're proud of

We built an end-to-end planning workflow with optimization, ML, simulation, visual comparison, and explainability. In historical replay, our v2 optimizer won 53/53 weekly windows and showed strong potential savings.

What we learned

We learned that industrial AI is not just about better models. The real value comes from combining optimization, domain rules, clean data, fair evaluation, and explanations that operators can actually use.

What's next for Quick2Plan

Next, we want to add richer calendar features to improve production-time prediction, make demand dropping margin-aware when capacity is tight, and extend the system toward predictive maintenance so Quick2Plan can schedule interventions before breakdowns happen.

Built With

• astro
• catboost
• fastapi
• langchain
• ml
• react