Production-grade ML system: time series forecasting pipeline → inventory decisions → FastAPI deployment → Docker containerisation.
| Skill | Implementation |
|---|---|
| Time Series | Naive, Moving Average, Seasonal Naive, SARIMA, Prophet, LSTM |
| Optimisation | EOQ, Safety Stock, Reorder Point (Operations Research) |
| ML Engineering | Model comparison, RMSE/MAPE evaluation, best-model selection |
| API Design | FastAPI with /forecast + /recommend_inventory endpoints |
| Deployment | Dockerised service, production-ready |
| Business Impact | Stockout reduction quantified in £ |
Retail Sales DB (SQLite)
│
▼
┌──────────────────────────────────┐
│ Forecasting Pipeline │
│ Naive → MA → Seasonal Naive │
│ → SARIMA → Prophet → LSTM │
│ (ranked by RMSE + MAPE) │
└────────────────┬─────────────────┘
│ Best Model Forecast
▼
┌──────────────────────────────────┐
│ Inventory Optimiser │
│ Safety Stock = Z × σ × √LT │
│ EOQ = √(2DS/hC) │
│ ROP = μ × LT + SS │
└────────────────┬─────────────────┘
│ Recommendations
▼
┌──────────────────────────────────┐
│ FastAPI Service (port 8000) │
│ POST /forecast │
│ POST /recommend_inventory │
│ GET /stockout_impact │
│ GET /model_comparison │
└──────────────────────────────────┘
│
▼
Docker Container
project2-demand-forecasting/
├── data/
│ ├── generate_data.py # 2-year retail sales (10 products × 5 stores)
│ ├── retail_sales.db # SQLite database
│ └── csv/ # CSV exports
├── models/
│ ├── forecasting.py # 6 forecasting models + comparison
│ ├── inventory_optimizer.py # EOQ + safety stock + ROP
│ ├── model_comparison.csv # RMSE/MAPE leaderboard
│ └── inventory_recommendations.csv
├── api/
│ └── main.py # FastAPI application
├── Dockerfile
└── requirements.txt
# 1. Install dependencies
pip install -r requirements.txt
# 2. Generate 2 years of retail data
python data/generate_data.py
# 3. Run model comparison
python models/forecasting.py
# 4. Generate inventory recommendations
python models/inventory_optimizer.py
# 5. Start API
uvicorn api.main:app --reload --port 8000
# Docs at: http://localhost:8000/docsdocker build -t demand-forecasting .
docker run -p 8000:8000 demand-forecasting| Table | Rows | Description |
|---|---|---|
sales |
36,500 | Daily sales: 10 products × 5 stores × 730 days |
inventory |
36,500 | Daily inventory levels + reorder events |
products |
10 | Price, cost, lead time, category |
Synthetic features:
- Seasonal Christmas peak (+40% in Dec/Jan)
- Weekend uplift (+35%)
- Upward trend (+0.02%/day)
- Promotional days (10% of days, +40% demand)
| Model | RMSE | MAPE |
|---|---|---|
| Moving Average (7d) | 9.84 | 14.9% |
| Naive Baseline | 9.89 | 15.9% |
| Seasonal Naive | 10.15 | 13.5% |
| SARIMA(1,1,1)×(1,1,1,7) | 10.43 | 16.7% |
| Prophet (optional) | — | — |
| LSTM (optional) | — | — |
| Product | Safety Stock | Reorder Point | EOQ | Annual Cost |
|---|---|---|---|---|
| Protein Powder | 49 units | 171 units | 704 units | £2,257 |
| Yoga Mat | 37 units | 140 units | 684 units | £1,440 |
| Wireless Headphones | 29 units | 127 units | 245 units | £2,323 |
# Forecast 30 days of demand
curl -X POST http://localhost:8000/forecast \
-H "Content-Type: application/json" \
-d '{"product_id": "p05", "store_id": "store_1", "horizon_days": 30}'
# Get inventory recommendation
curl -X POST http://localhost:8000/recommend_inventory \
-H "Content-Type: application/json" \
-d '{"product_id": "p05"}'
# Business impact report
curl http://localhost:8000/stockout_impact"Built and deployed demand forecasting system (SARIMA + Moving Average) with FastAPI + Docker. Implemented inventory optimisation using EOQ and safety stock models, projecting 18–30% stockout reduction and measurable holding cost savings across 10 product lines."