Built on published research, not vibes
Kairos's clip selection is driven by BRAID — a schema-generation + executor architecture published in our research line. We don't share the secret sauce, but we'll show you the results.
BRAID decision engine
Two-stage system: a structured schema generated for the clip's domain, then executed by GPT-4.1 to score moments.
Semantic signal processing
Audio + transcript + visual signals fused into a single salience score per second. No black-box vibes-based ranking.
Result transparency
Every clip ships with a virality score breakdown — emotional, standalone, shareability, platform-fit.
How to use Research
- 1
Schema generation
BRAID inspects the source video and proposes a content-specific schema for what makes a great clip in that domain.
- 2
Executor pass
GPT-4.1 evaluates every candidate moment against the schema, returning a structured score per dimension.
- 3
Multi-modal fusion
Signal-processing layer integrates pitch, energy, and visual cuts into the schema score for context.
- 4
Output: explainable picks
Every chosen clip has a reason and a breakdown — never an opaque "trust us".
Frequently asked
It's our core competitive moat. We publish results (benchmarks, virality scores, comparisons) — not the prompts or schema generators.
GPT-4.1 with a structured-output schema. We chose OpenAI for executor quality on Turkish + English bilingual content.
Internal research papers; we publish technical reports and benchmarks. Not academically peer-reviewed yet.