← Back to work

Case Study · 2024

AI-Powered Demand Forecasting

TensorFlow models forecast product demand from sales and inventory history — served through Django REST and a React dashboard.

RoleSoftware Engineer II
TimelineTeton Private Ltd.
FocusTensorFlow · Django · React
TF

Context

Ops had sales and inventory history, not a forecast. They needed a model, an API, and a screen.

The Problem

Without a forecast, restocking is a gut call. Gut calls don't scale across a catalogue.

Process

01

Prepare the series

Python data-preparation pipelines from historical sales and inventory.

02

Infer behind REST

TensorFlow inference services exposed through Django REST for downstream apps.

03

Make the forecast readable

React dashboard for demand trends, model outputs, and product-level insights.

Decisions

The model is not the product

Without an API and a dashboard, a notebook stays a notebook. Inference had to live in Django.

The grain is the product

A global forecast doesn't help the buyer. Each SKU gets its own curve.

Outcome

Demand forecasts served in production via Django REST

React dashboard for trends and product-level insights

Reflection

A model in production is not an .h5 file — it's an API someone opens on Monday morning.