Sikshana Foundation meeting · Friday 30 October 2026 · working demo, offline

Sikshanapedia + Aural: listen, then ask

Sikshana already reaches children through Sikshanapedia and Anganwadi workers through Shishu Vikasana's daily audio instructions. Both are mostly one-way. Aural adds the other half: the learner asks out loud and gets an answer taken only from the lesson they just heard. Together, every Sikshana lesson becomes something a child or a worker can listen to and then ask about, in English or Kannada, with no internet, on the same school laptops and Raspberry Pi as Sikshana Spoken English.

What the demo does

  1. Who is learning? Children, or Anganwadi workers.
  2. Pick a topic. 26 topics, filtered by subject: 17 for children (Science, Environment, Health, Maths, Stories: Why Do Bees Buzz?, How Do Aeroplanes Fly?, Mosquitoes…) and 9 for workers (My Balwadi, Happy and Sad, Too Many Bananas…).
  3. Listen. Each page has a picture, English and ಕನ್ನಡ text, and a recorded clip in each language. Read all plays the whole topic.
  4. Ask. Tap Ask and speak. The answer is spoken, shown and linked to its page, which plays in Kannada too.
“Why do mosquitoes bite?”The female mosquito needs blood to make her eggs. (page 3)
“Who won the cricket match?”I could not find that in this topic. Try asking about something in it.
The topic screen for Mosquitoes: the page with its picture, English and Kannada text and Listen buttons on the left; on the right the Ask button, the question, the answer with Page 5, Again and Kannada buttons, and suggested questions.
The topic screen on the Pi's touchscreen (also runs in a window on a laptop).

Why it is safe for children

  • Answers only from the lesson. A question is matched to a page of the topic. A small open model may say that page in a shorter sentence, but its answer is dropped if it uses words that are not on the page, and the page itself is read instead.
  • Voices are never kept. The question is written down on the device and the recording is dropped. Nothing is sent anywhere.
  • Open-source models only, the same engines as Sikshana Spoken English. An answer takes a few seconds.

About the content

Sikshana's own content is not public, so the demo uses open material in its place. Kannada for the non-Pratham topics is a draft that needs review. One Wikipedia page we used had been vandalised, which is why every lesson needs a human review before children see it, as in Aural.

SourceUsed forLicence
Pratham Books StoryWeaver22 stories with their published Kannada; picturesCC BY 4.0
Simple English WikipediaThe Moon, Mosquitoes, Clean handsCC BY-SA 4.0
A2Z, for the demoCounting activity, talk ideas, Kannada draftsCC BY 4.0

What we would need from Sikshana

Real contentA few Sikshanapedia lessons per age group and a week of Shishu Vikasana audio packs.
A reviewerSomeone who reads each lesson and its Kannada before it goes on the devices.
Real questionsWhat children and workers actually ask, from a short pilot, to link common questions to pages.

Next steps we propose: load 10-20 real Sikshana lessons with their own audio; let learners ask in Kannada (an open Kannada speech model such as AI4Bharat IndicConformer); a 2-week pilot in 2 schools and 3 Anganwadis on laptops already there, plus one Pi.

Models: Whisper (speech), Qwen2.5 (answers), AI4Bharat Indic Parler-TTS (English and Kannada clips), Kokoro/Piper (live voice); all open source, all on the device. Full list of sources and licences: docs/pedia/sources.md. Stories and illustrations: Pratham Books StoryWeaver (CC BY 4.0). Text adapted from Simple English Wikipedia (CC BY-SA 4.0). Moon photo: Gregory H. Revera (CC BY-SA 3.0). Mosquito photo: CDC (public domain).