Major project IT, Software & SaaS Case Studio

Personalised thumbnail chooser with contextual bandits

Inspired by publicly reported work at Netflix

Choose which image to show each viewer for a title, learning from clicks with a contextual bandit, and test it offline.

The brief

Netflix has described choosing a different artwork for the same title for different members, treating it as a contextual bandit: show an image, see whether the member plays, and learn. Build the same kind of system on open data. Train bandit policies on logged recommendation data, evaluate them offline with off-policy estimators, then build a small service that picks a thumbnail for a viewer and learns from feedback. Good looks like: a correct off-policy evaluation (IPS, doubly robust) against the logging policy, a running demo that adapts, and an honest note on bias and what you could not measure.

Read the case first

What Netflix has said in public about this kind of work, on its own pages:

Make it your own

Change what is being chosen or for whom: banners for an online course catalogue, covers for an Indian-language ebook store, or the order of rows on a home screen. Write down what the bandit is allowed to learn about the viewer and what it must not use.

AITeky has no link with Netflix, which has not reviewed or endorsed this brief. The brief only draws on what the company has published, uses open data rather than its data, and is a learning exercise: don't present your project as Netflix's work or system.

What you will build

Study the logged bandit data: actions, context and propensities
Implement epsilon-greedy, LinUCB and Thompson sampling policies
Evaluate policies offline with IPS and doubly robust estimators
Build image features for titles (colours, faces, text) from posters
Serve a thumbnail chooser API that logs propensities and learns online
Simulate viewers to show the policy adapting, and report fairness and filter-bubble risks

Free, open data to build it with

It takes a minute with your email. AiTeky Pass (₹999/month or ₹9,999/year) is free for your first 30 days.