An EPFL startup
AI decision intelligence
for broadcast programming.
Choi helps programmers choose what to show, on every channel and platform.
Its own model learns from your data and your teams’ choices, on your own infrastructure.
Saturday night box office.
Which film airs?
Thousands of titles could fill the slot.
Choi narrows them down to a few strong options.
- 4,812 titles in the catalogue
- 1,140 could air that night
- 5 suggestions from Choi
- Your programmer chooses one
Illustrative figures.
- Inception
- La La Land
- Amélie
- The Intouchables
- Bohemian Rhapsody
One centralised cockpit for the complete offer.
Choi gathers your programming data in one place and puts it to work wherever your teams choose what people watch.
Trending now
Documentaries
Your data, augmented
Choi works from the catalogue, contracts and schedules you already have. It makes them easier to use, so you get more from a catalogue you’ve already paid for. And they stay yours.
Sovereignty by design
Choi runs on your own infrastructure, alongside the tools your teams already use. You keep control of your data, and of everything the model learns from it.
A system that learns from your choices.
- Choose what to fill:
Friday night box office, Thursday morning “Classics” or the streaming home page? - Choi suggests titles from the whole catalogue, adapting to the context, with built-in explainability.
- Your programmer chooses from the top suggestions.
- Choi adjusts its model parameters, choice after choice, so the next suggestions fit your needs better.
Try it out!
Each broadcaster chooses its own objectives.
Tuesday, 20:10
IllustrativeThe model’s parameters
Choose a title. Choi shifts its model parameters toward what that choice valued.
Our own model.
Peer-reviewed.
Choi runs on a model we designed ourselves. Not a chatbot, not a language model: context-aware reinforcement learning from human feedback, built for programming decisions and weighing several goals at once.
Context-aware
Saturday night box office and Sunday morning oldies call for very different decisions. The suggestions adapt to the context of each slot.
Multi-objective
Several objectives at once, set by each broadcaster. Choi learns how to balance them from past programming and from your teams’ choices.
Human in the loop
Your programmers make every decision.
Each choice is the feedback the model learns from, so it keeps adapting to how your teams work.
Explainability
Every suggestion shows the signals behind it, so your teams can make an informed decision.
News
- Choi at the EPFL Investor Day.
- The method is presented at ICPR 2026 in Lyon.
- The paper is published on arXiv.
Questions
What data does it use?
The data you already have: your catalogue and content management system, your schedules, and your audience analytics, on air and digital. Choi builds on it and enriches it with external data.
Where does our data go?
Nowhere. Choi runs on your own infrastructure, so your data, and everything the model learns from it, stays with you.
Is it compatible with our tools?
Yes. Choi works alongside your systems and is connected to them during setup. Your teams keep the tools they are familiar with.
How soon can our teams use it?
As soon as setup and integration are done. From then on, Choi is up and running for your programmers to use in their daily work.
How do we get started?
Write to us at hello@choisystems.com. We start with a conversation about your channels, platforms and goals, then set Choi up on your infrastructure, connected to your data and tuned to your editorial identity.
Let’s talk.
hello@choisystems.comWe work with a few broadcasters at a time. Write to us and we’ll reply as soon as possible.