Ask a question, get an answer from the source. Nalanda is a learning app that answers your questions with passages from ancient texts, shown alongside their original context.
I built it on my own, from idea to App Store: the roadmap, the design, the SwiftUI and Supabase engineering, the RAG search, and the launch. I also set up the business side, from forming the LLC to writing the terms of service.
I'm also in the competition's official highlights reel.
Click through the slides. The animations play just like they did on stage.
From one book on a laptop to a library on the App Store, and the problems I had to solve along the way.
Nalanda began as a desktop app for a single book, the Bhagavad Gita. I built it in Electron so I could read a verse in Sanskrit, see it word by word, compare translations and commentaries side by side, and listen to each one. By the end of Chapter 1 it had 15 verses, 54 recorded tracks, a dictionary of the people in the text, and a notes system.
While building it, I kept finding more: translations and commentaries of the Gita, the Mahabharata, and other Eastern texts, published long ago and now in the public domain. Most were sitting unused as scanned PDFs. One book was too small a scope.
To reach the most people, I also had to leave the desktop. People carry their phones everywhere, so I rebuilt Nalanda as an iPhone app in SwiftUI.
These books were printed generations ago, and the scans show it: faded type, broken letters, and stray marks. Many pages switch between Sanskrit in Devanagari script, transliteration with accent marks, and English, sometimes within a single line. When the software misreads a word, someone has to work out what it was meant to say.
I tried nearly every major OCR service out there. None could read these pages reliably.
I got free GPU time through a student plan, which meant I could run serious AI models myself. I searched through open-source models for ones that could handle these pages, and found that no single approach worked for every book: each one needed its own pipeline, running several models over multiple rounds.
The most powerful GPUs were usually busy during the day, so I ran the jobs overnight, night after night, for weeks. That's how these books became clean, searchable text.
With the text in hand, I wanted more than a reader. For questions this important, I didn't think an AI's paraphrase was good enough. I wanted Nalanda to work like a compass: point you to the verses and passages that speak to your question, so you can read them yourself and reach your own conclusions.
Around then I was taking vector calculus, and I wondered: if vectors can hold that much information, could they hold the meaning of text? That became Nalanda's search. The Gita was simple, with one verse to one vector. Other books aren't written in verses, so each one needed its own way of being broken up and turned into vectors.
I tested five embedding services and different vector sizes until the search returned what a careful reader would point to. The service that turns the books into vectors also has to turn every question into one, so that choice had to be right.
At first, to keep costs down, I turned questions into vectors on the iPhone itself. It was slow, and older phones couldn't handle it, so I moved that step to the cloud.
The books moved too. Shipping them inside the app would have made it enormous and easy to pirate, so the texts live in a Supabase database instead, and the app is under 15 MB. I then tuned what to cache and when, cutting home-screen database work 30×. Questions are stored anonymously, so a question someone has asked before is answered instantly instead of being computed again. That keeps AI costs around $1 per user per month.
Not everyone wants answers from every book, so you can choose which books the Guru searches. I added profiles, following and public playlists so people can share what they're reading, along with the moderation every social app needs: reporting, blocking, and content guidelines. A home-screen widget, daily reminders and notifications for new content bring people back.
Then came beta testing. I kept it to about half a dozen testers, who signed up through a form on nalandaguru.com, so I could act on every piece of feedback before launch.
I hired a registered agent, formed an LLC, and opened a business bank account. Then I wrote the privacy policy and terms of service, and registered a DMCA agent.
Then there was the question of payment. Apple's in-app purchases take up to 30%. As a longtime Spotify user, I knew some apps send you to their website to subscribe, and US App Store rules allow that link. So I built checkout on Stripe and tested it in Stripe's sandbox. When someone uses up their free Guru questions, Nalanda invites them to subscribe on the web.
Apple rejected the first submission. I emailed, I called, I fixed what they asked for, and I resubmitted. In the meantime I built nalandaguru.com. On October 3, 2026, Nalanda was approved and went live on the App Store.
I also pitched it at the Santa Cruz Launchpad startup competition, and won $500 at the finals.
Here's how a session in the iPhone app flows, from opening it to saving what you found. The screenshots are from nalandaguru.com. Click any of them to enlarge it.
Home is built to get you reading in one tap. The top row picks up the Gita chapters, each with its own cover art. Below that, shelves suggest what to read next: collections built around themes like Transmuting Grief Into Purpose, and picks made for you based on what you've read.
Each card shows how much it holds, for example 1 verse · 14 passages.
This is the heart of Nalanda. Ask a question in your own words, and the Guru answers with real passages from the source texts instead of an AI's paraphrase. Each result names the book, the section and the part it came from, so you can read it in context.
Behind it is the retrieval search I built across texts that I digitized with my own OCR pipeline. I designed it to cost about $1 in AI per user each month.
Tap the bookmark to save an answer to your Library.
Open any verse to read it in the original Sanskrit with its English translation. The tabs switch between both together, the Sanskrit alone, or the translation alone. Swipe left or right to move through the chapter.
The Verse tab also plays the chanted recording.
Search covers every book, chapter and passage in the app. If you'd rather browse, each book is laid out as a shelf of its parts: the Gita by chapter, and the Mahabharata by parva, each with its own artwork.
Use Filter by Book to search one text, or Filter for Friends to find people you know.
Everything you save lands in your Library: Guru answers, passages, verses and your own notes. Pin the ones you come back to, sort them into folders, and build playlists of passages around a theme, like Overcoming Grief or Duty and Action.
Filter by Mine, Guru, Other or Notes.
Alongside the iPhone app, I've built Nalanda as a native Android app, and the Bhagavad Gita desktop app where it all started.
After launching Nalanda on iOS, I rebuilt it as a native Android app. It isn't a wrapper: it's a full rewrite in Kotlin and Jetpack Compose, ported from the Swift app.
The rule for the port was that the numbers come from the Swift, not from Material Design. Card sizes, drawer width, colors and the type scale were all copied from the iOS code, so the Android app feels like Nalanda rather than a stock Android app.
It compiles with zero warnings and runs end to end on the Android emulator against the live backend: Home, Library, Guru, Discover, Search and the readers.
The code compiled cleanly, but running it on a real emulator turned up five bugs. Four of them would have been hard to catch in code review alone.
The desktop app is a focused study tool for the Bhagavad Gita. Each verse is layered: the Sanskrit, a romanized version and word-by-word breakdown, an English translation, and two commentaries. Every layer has its own recorded audio.
Nalanda is live on the iPhone. Here's where it's headed, and where each version stands today.
The full product: ask a question and get the passage from an ancient text that answers it.
A native rewrite of the iOS app that runs end to end on the Android emulator.
A Bhagavad Gita study desk with layered verses, recorded audio and notes.
Nalanda is on the App Store for iPhone. If you'd like a demo of the desktop or Android builds, send me an email.