AI disclosure

Last updated 3 May 2026

KathaSagaram is an AI-told, human-curated, openly-correctable retelling of Indian sacred stories. We believe families deserve to know exactly how the stories their children hear are made — so this page tells you the entire process, including where AI is and isn't involved.

How a story is made

1 — Source selection

A human (currently the founder, Sripad) picks the story from widely-loved retellings in the public-domain Indian canon — Bhagavata Purana, Ramayana, Mahabharata, regional Puranas. The source text is cited on each story's page.

2 — Script drafting

A large language model drafts a 2–3 minute kid-friendly script in English, working from the source. The draft is then translated into Hindi and Telugu by the same model with explicit checks for proper nouns and pronunciation hints from a hand-curated lexicon.

3 — Human review

A human reviewer reads the full script in all three languages against a written rubric (cultural accuracy, age-appropriateness, narrative warmth, theological care, regional sensitivity). Each language must score ≥32/40 on the rubric. Nothing publishes without a human signoff.

4 — Voice synthesis

An AI text-to-speech voice records the script in each language. The voice is one of a small number of locked, consented voices we license from our TTS provider. We do not clone real people's voices.

5 — Illustration

The hero illustration is generated with an AI image model using a fixed style anchor (for visual consistency across the catalog) and per-story prompts. The reviewer checks every illustration for cultural accuracy — clothing, jewelry, divine attributes, regional details.

6 — Quiz & talk-prompts

The five-question Mokshapatam quiz and up to seven "talk-about-it" prompts are also AI-drafted and human-reviewed against the rubric.

What's not AI-generated

The choice of which stories to tell, in what order.
Pronunciation lexicons (curated by hand from speakers of each language).
The cultural rubric used to score every script.
Every published version's signoff.
This disclosure, the privacy policy, and the FAQs.

How we handle mistakes

We will make mistakes. AI models hallucinate; cultural details vary by region and tradition; pronunciations carry weight. Three things we built into the app from day one to handle that:

Versioning. Every story carries a version number and a "replaces version" pointer. When we ship a correction, the old version is retired but its analytics history is preserved so we can learn from what changed.
Retraction kill-switch. A single command — retract.py story_id=S0001 — pulls a story from every device within five minutes. We test this end-to-end before every release.
Public correction loop. A small feedback button on every story and character opens a pre-filled WhatsApp message to +91 9990912205. We aim to acknowledge cultural and pronunciation concerns within 24–48 hours and ship a fix within 7 days. We post correction notes inside the app via the AppUpdateBanner.

Cultural advisor cycle

Starting on Day 7 after launch, every story is sent to an external cultural advisor with a 7-day SLA for substantive feedback. Advisor signoff is a marketing gate (we don't claim formal scholarly endorsement until they sign off), not a ship gate.

The models we use

For transparency, the providers powering the app today:

Script drafting + translation: Anthropic Claude (large language model).
Voice synthesis: ElevenLabs Multilingual v2 (text-to-speech).
Illustration: Midjourney via PiAPI, with a locked style anchor.
Quality evaluation: Claude (script grader), human reviewer (final).

We re-evaluate this stack quarterly. If a model changes meaningfully, we note it here.

What we tell your child

Inside the app, in Settings → About, every child sees: "These stories are told by a computer, then checked by a person. If you hear something funny, ask a grown-up to tell us."

We believe that's the right level of disclosure for a 4–8-year-old: honest, direct, and an invitation to talk.

Open to scrutiny

Spotted a problem with this disclosure, or a story you'd like us to look at? Message us on WhatsApp or email hello@kathasagaram.app. We read everything, and we publish substantive corrections.