◆ Kuku × Dvar · Concept demo

Every story checked.
Every model measured.

Kuku already runs AI across its pipeline: scripts, dialogue, dubs, thumbnails, descriptions. Dvar is the measurement layer under it: checks on every scene, evals for search, drafts and dubs by language and genre, and one gateway that sends each call to the cheapest model that clears the bar.

Start the tour Open the live studio

Cost per correct outcome, Dvar against the frontier baseline

4.5×

Content checks: lower ₹ per correct scene

₹0.03 vs ₹0.12 · 96% vs 93% right

36×

Search: lower ₹ per correct search

₹0.0022 vs ₹0.08 · 100% vs 100% right

4.5×

Episode drafts: lower ₹ per correct draft

₹0.27 vs ₹1.24 · 67% vs 25% right

1.4×

Localisation: lower ₹ per correct dub

₹0.10 vs ₹0.14 · 67% vs 100% right

Placeholder run, not measured: these figures show the method working, not results for any model. Live studio calls are measured.

Ten invented shows on the test bench

Not Kuku shows

Try it now

Live studio

Check a scene, read a search, write an episode or dub a scene. An open model on Dvar and a frontier model, side by side, with cost and time.

For the AI team

Evals bench

Every model on every task, ranked by how often it is right and what a right answer costs, split by language and genre.

For content ops and standards

Content checks

Foul language for the rating, mixed language, mixed context, residue and content above the rating. Code first, a small open model next, frontier only when unsure.

For search and discovery

Search evals

Titles, misspellings, Devanagari and “bhoot wali kahani”, scored by query type, with the model step's cost and p95 latency.

For the storytelling team

Generation evals

Every generator update read against episode briefs. One critical failure blocks; cost per accepted draft decides.

For dubbing

Localisation evals

Hindi scenes into Tamil, Telugu, Bengali and Marathi: script, names, Hindi or English left in, timing, meaning and tone.

For platform and finance

One gateway

Open and frontier models behind one endpoint, each workflow on the cheapest model that clears its eval, with fallback.

What Kuku has said in public

Generative AI at every layer: adapting stories to the Indian context, scripts, dialogue, thumbnails and descriptions, with custom GPTs for each process.

Analytics India Magazine, Feb 2024

Genre-specific model routing, self-reported by the CEO: certain models handle science fiction, others true crime. How that ranking is measured is not public.

Financial Express, Oct 2024

KukuTV: 200–250 shows a month, targeting 1,000. AI for story-potential scoring, script fixing, pre-visualisation, hook and cliffhanger identification, VFX, dubbing and music; dramatic language in Indian languages still not good enough for end-to-end dubbing.

THR India, Jun 2026

English writing moved fully to AI; Hindi and Tamil quality described as decent; Telugu, Kannada, Malayalam, Bangla and Marathi called still poor (dated).