Writing on AI tutoring and personalization.
What we're learning as we build GyanSpark — the ideas, the trade-offs, and the occasional thing we got wrong.
Storing student data responsibly
Building a memory layer means deciding, deliberately, what's worth remembering about a child's learning — and what isn't. A few principles we hold ourselves to while we finish writing the formal policy.
Read article →Personalizing flashcards, not just chat
Most "AI memory" demos are chat demos. The same idea applies just as directly to flashcards — arguably more usefully, since wasted review time is where the cost is easiest to see.
Read article →Memory vs. context window vs. RAG
These three get used almost interchangeably, and they solve different problems. Knowing which one you actually need saves a lot of wasted engineering.
Read article →Why AI tutors forget
Every AI tutoring feature has the same blind spot — it only ever sees the conversation in front of it. Here's why that's a harder problem than it sounds, and what it costs.
Read article →