Five people play. Everyone else watches.
The NGO has spread Jain philosophy for over 15 years, and quiz games are one of the ways it teaches. But at every event the format was the same: 4–5 people on stage answered questions, and the rest of the hall sat and watched.
People remember what they do far better than what they watch. A quiz you only watch is just a lecture with a scoreboard.
Put the game in every pocket.
Everyone in the hall already had a phone in their hand. So instead of a stage game, every round is played on phones, with the projector showing the question to the room.
आज कल सब वैसे ही मोबाईल में लागे रहते हैं आपने गेम्स को मोबाईल में डाल कर उसका अच्छा सदुपयोग कराया"Everyone is on their phone anyway. Putting the games on the phone made good use of it."
Joining had to cost nothing: no app, no account, no password. A code, a name, and you're playing, in Hindi or English.
One host, one laptop, three screens in sync.
It grew into a platform: a game Studio with six game types, several venues at once, paper-and-phone tambola, Hindi and English, and an offline rehearsal mode. I built it with Claude as my engineering partner. The NGO has no tech budget, so it runs entirely on free services.
A host on stage is already talking, timing and reading the room. The controls had to need almost no thought.
The hall played.
अब हम लोग बोर नहीं हो रहे क्योंकि हम भी खेल पा रहे हैं इन गेम्स को।"We aren't bored any more, because we can play these games too."
आपने स्टेज और ऑडियंस दोनों के लिए खेल का पार्ट बना दिया अच्छा लगा"You gave both the stage and the audience a part in the game."
What players typed vs. what counted
The scoreboard said 38%. It was wrong.
That show used typed answers. I went through every answer the system had marked wrong.
Nothing drains motivation faster than losing points you earned, and here it hit exactly the people we most wanted to include.
Some got bored. Nobody learned why.
आज की जनरेशन के लिए तो अच्छा है पर हम जैसे बूढ़े लोग थोड़ा बोर हो गए।"Good for today's generation, but older people like us got a little bored."
The data agreed. Two of the seven questions that night were answered correctly by just 1 person each, while another was answered correctly by 26 of 41. And after each reveal the phone said only right or wrong: 0 of 15 questions in that game had an explanation.
Too hard and people give up; too easy and their attention drifts. A hall aged 8 to 75 needs more than one difficulty, and the "why" after a reveal is where the teaching happens.
Four frictions, one root
me, the night before, ~1 hour by hand
One root under all of it.
Three separate fixes would cost three times the effort. Changing how questions get written fixes all of them at once.
What if I reviewed questions instead of writing them?
Medium · PDF page 47 · with an explanation to read on stage
Be the player with no Hindi keyboard
"Ajeev" (अजीव) stays wrong. It means the opposite, so being tolerant must never mean being loose.
Judging a question is much easier than inventing one. The AI drafts; I recognise the good ones, fix or drop the rest, and nothing is added until I tick it.
Version 1 of the safety check
✓ Found in the book
Version 2: the blind check
✗ Disagreed, left unticked
My first safety check agreed with nonsense.
This is religious teaching in front of children and elders, so a wrong "fact" on the projector is a real harm, not a typo. I added a second AI to check every draft. It approved a riddle whose two clues pointed to two different words.
A reviewer who sees the answer first is looking for reasons to agree. Take the answer away and you get an independent opinion.
What if the AI isn't there?
In testing, Google's four newest models all replied "high demand", and one request hung for over three minutes. A show can't depend on someone else's server.
When a part fails, the product should get a little worse, not stop. The manual path never depends on the AI. (The screenshot is from a deliberate outage test.)
North Star
Honest status, and how I'll know.
The AI feature has been live as a Beta in the Studio since 14 October 2026. It has not yet been used in a real show, so the numbers on the left are targets, not results. Next: three shows where every hall plays one AI round and one hand-written round, with my prep time measured, and a scholar checking every AI question before it reaches the projector.
Count what reaches the hall and teaches something, not how many drafts the AI produces. Otherwise the metric rewards volume nobody uses.