Kalezio
Built for Hollywood, ready for every regional market — simulate 10,000 AI audience reactions to know how a film will land before you shoot it.
The Challenge
Studios and independent producers commit millions to a script before ever finding out how a real audience will react to it. Test screenings happen after the film is shot and cut — far too late to change course cheaply.
See Your Audience Before You Shoot
A test screening tells a studio how an audience felt about a film that's already been shot, cut, and paid for. Kalezio moves that signal earlier — to the script stage, before the expensive decisions get made.
One Action, One Verdict
The product is deliberately narrow: upload a screenplay, run a screening, get a verdict. That single POST /screenings action replaced an earlier, more complex multi-step pipeline — fewer steps between a script and an answer.
A Simulation Engine, Not a Single Prompt
Asking one LLM call "how would an audience feel about this script?" is a guess. Kalezio's simulation engine runs three rounds — archetype clustering, opinion dynamics, then consensus — batching thousands of simulated agents into a manageable number of LLM calls, with context caching keeping the cost of each screening down without cutting the number of simulated reactions.
Recognition-Aware by Default
A known title, cast, or IP changes how a real audience reacts — so by default, Kalezio's simulated audience reacts to that context too, the way a real one would. Producers who want a blind read can strip that identifying metadata and see how the script performs on its own.
Our Solution
Kalezio runs a virtual audience screening on a screenplay before a single scene is filmed. Writers and producers upload a script, run a screening against a simulated audience of up to 10,000 AI agents, and get back a verdict — grounded in a three-round in-house simulation engine, not a single LLM prompt pretending to be an audience.
Tech Stack
Results
- Simulates up to 10,000 AI audience reactions per screening
- Three-round simulation engine — archetype clustering, opinion dynamics, then consensus — not a single-shot AI guess
- Context caching cuts the LLM token cost of a screening by roughly 75% without reducing the number of simulated agents
- Every screening failure refunds credits automatically — no silent failures the customer pays for