Designing an AI that speaks Hinglish
A code-mixed peer voice across three age tiers, gentler for Class 6 to 8 and more relaxed for 11 to 12. We answer in the language a student already thinks in.
Field notes
How we think about building AI for students.
Working notes on what students need, and how we know when we're getting it right.
Featured note · Thesis
More content was never the bottleneck.
Students already drown in decks, PDFs and lectures. What they don't have is someone to ask at 11pm in their own language. That is the gap Infi closes: a Hinglish companion that answers the doubt right now and turns a low moment into one useful next step.
The conviction underneath everything
Open Infi, get help, do the next thing, come back. When that loop holds, the thesis holds. Every other note here sits downstream of it.
The notes
Points of view, in progress.
Positions we keep pressure-testing as we ship. How we reason, not doctrine we defend.
Every answer ends in a next step
A solved doubt is a dead end unless it points somewhere, so every reply closes with a next action.
Measuring real learning, not vanity
Our north star is Weekly Active Learners: three or more real value actions in seven days. A login looks fine on a chart. Coming back to do something useful is what actually counts.
Guardrails for minors: redirect, don't reject
Most of our users are children, so we redirect with humour rather than lecture, and we never shame. The one hard line: anything touching self-harm goes straight to real helplines.
Feedback as a system
Beta calls, in-app behaviour and analytics feed one opportunity solution tree, so an insight becomes something we can act on instead of an anecdote.
Public good
AI for public good: the Poshan AI approach
AI should earn its place by helping the people the system misses. Poshan AI flags child malnutrition early, built for fairness and run with the state.
How we research
The principles underneath the notes.
Four habits that keep our points of view honest as the product moves.
01
Start from the student, not the slide
Start from the student, not the slide. Every note begins with a real moment: a doubt at night, a parent hesitating, a student about to give up. We watch first and theorise after.
02
Measure what counts as a win
Measure what counts as a win. We decide the value action before we build. If a feature can't move a real learning metric, it doesn't earn the screen.
03
Safety is a design input, not a patch
Safety is a design input, not a patch. When you build for minors, the guardrails come first: tone, redirection, escalation, decided up front rather than bolted on after something breaks.
04
Hold every position loosely
Hold every position loosely. These are field notes, not commandments. The moment a student shows us otherwise, the note changes. We would rather be corrected fast than right slowly.
Work on this with us
These problems are open. Come solve them.
Hinglish-native AI, real learning metrics, AI for public good. If this is how you'd want to spend your days, we should talk.