Quick look at Claude Code-generated Bhagavad Gita React Native and Expo app (no coding by author)

Last updated on 11 Feb. 2026
Recently a Bhagavad Gita AI generated iOS app with no coding required from app creator was covered in Indian news media outlets. Yesterday, 7 Feb. 2026, I had a look at one such media article which led to me cloning the app's GitHub repo and run the app on dev server for web and Android.


It linked to the X post on the iOS app: https://x.com/AnishA_Moonka/status/2019321410699817458, Feb 5, 2026. The post said:
  • Author only described what he wanted in plain English to Claude Code
  • Claude Code 'wrote everything' (developed the app).
  • Codex reviewed the app.
  • $200 Claude Max subscription
  • $20 ChatGPT Pro subscription
  • $99 Apple Developer fee
  • App is called 10minutegita
I don't have an iOS device and so I could not try out the app.

I had read somewhere that the app was a React Native and Expo app and IFIRC that it was Open Source.


I decided to have a quick look at the repo. ...

While the source code has been shared, it is released under a restrictive license that does not permit free reuse. As such, it cannot be considered open source in the usual sense, unlike projects released under permissive licenses such as MIT.

The Bhagavad Gita data it uses is baked into the app as JSON files and that data is also proprietary. This data seems to be associated with Swami Chinmayananda (explained later in this post).

I cloned the repo and tried it out. Some notes about that:
  • npm install needed audit fix for 2 vulnerabilities. After that, 0 vulnerabilities.
  • npm run start: gave some warnings but app showed up on web - http://localhost:8081/
  • Vertical scrollbar appeared for many paragraphs if window width is reduced to mobile size (or even larger than mobile size).
  • The bottom of the Gita page contents app has an attribution notice - "Inspired by the teachings of Swami Chinmayananda".
  • The app has a fixed viewing structure - verses per day as per its selection.
  • One cannot go to next day without marking earlier day as complete.
  • No choice to go to specific chapter or verse.
  • Even after marking a day as complete, it allows user to only see preview of the next day. Beyond that, data seems to be completely locked. Quite strange!
  • The repo does not seem to have the Claude (AI tool) prompts used. That's a big missing thing here, IMHO.
  • The git commits mention Swami Chinmayananda as the author. The commit messages are descriptive.
  • Ran it on my Samsung M21 Android phone using Expo Go. A warning/error was shown about Android push notifications of some package removed from Expo Go.
  • The Android app seems to work in Expo Go - similar to web app but better as paragraph vertical scroll issue is not there.
  • The repo includes Jest-based automated tests under a __tests__ directory. These may be standard React/React Native unit or snapshot tests, running via npm test. They validate code-level behavior rather than real user interactions (unlike Playwright-style end-to-end tests).

Closing thoughts

If Claude AI prompts used were provided and linked to associated commits in the repo, it would have given a clearer picture of what was done by Claude AI, especially if Claude AI tripped up on some points and the author had to fix it with additional prompts.

Codex review contributions are not described. It would have been great to see what Codex review pointed out and how Claude fixed it.

Did Claude do all the testing using Jest? Was there additional testing done manually by using app in simulator and real device? I think testing is a critical part of any software lifecycle, whether for web or mobile applications. While unit testing provides useful validation at the code level, it cannot replace end-to-end testing that exercises real user actions and app flows. In the case of this Bhagavad Gita app, the testing visible in the repository appears limited to unit-level Jest tests and does not include end-to-end user interaction testing. This makes manual testing on simulators and real devices an essential step before any release, as otherwise there is a significant risk of user-visible and potentially embarrassing bugs. More: ChatGPT exchange on Testing of an AI generated app.

I just saw an interesting video about using Claude to generate a React Native and Expo math learning app: Claude Code Tutorial for Beginners: Build App with AI (2026), ~22 mins by Mikey No Code, 25 Jan 2026. The author clearly describes the prompts used. It would have been really great if the 10minutegita author had also shared his prompts.

The app’s feature set is relatively limited when compared to a general-purpose Bhagavad Gita app. It does not provide the ability to navigate directly to a specific chapter and verse, nor does it offer a choice of multiple translators or commentators. Instead, the app structures the presentation of Gita verses around a time-based progression, limiting what is shown to the current day, the next day (preview), and possibly past days. This time-oriented structuring may be a deliberate design choice by the app’s author, who may not have intended the app to support free navigation across chapters and verses.

Even if the app creator would not have done any coding, he may have been advised about what technologies to ask Claude to use - React Native and Expo. 

The generated code is quite readable and the commit messages are very descriptive. That is a strong positive feature as if manual technical review/update becomes necesssary for bug fixes or incorporating some features, that may be possible without much difficulty.

Overall, it is impressive to see what Claude Code can generate. Unfortunately, Claude Code does not seem to have a free tier. In my initial visit to Claude site, I had misread the Claude free tier as including Claude Code. Free tier has only the chatbot. I had planned to try Claude Code out by doing a 'prompt-along', if I may use that term, with above mentioned Claude Code tutorial video but that will have to be cancelled now, I guess. Related ChatGPT chat extracts: Impractical for me: Using Claude code for free with Ollama and free model.

I thank the 10minutegita app author for posting about the app and for sharing the source code of the app publicly on GitHub.

==================================================================
P.S. Some YT videos I saw in the context of this post.
  • Dario Amodei speaks quite frankly. I could see the scientist background clearly which was unexpected. I had expected a typical CEO talk.
  • Amodei clearly says that Anthropic focuses on enterprise customers. That explains why Claude Code or Cowork do not seem to have a free tier.
  • Amodei's ethical concerns and long-term views about AI are interesting.
First Look: Claude Cowork Is the Real Deal, ~16 mins, Jan. 2026, by Matt Maher
  • I saw only till the first 10 mins of the video which covers: "DEMO 1: Mythology Notes". The author pointed Cowork to a folder with lots of class notes including PDFs and handwritten notes. Based on prompts, Cowork then understands the contents of the folder, provides a detailed answer to a specific question about it, and builds a progressive study guide. It seems Claude does not generate images but the author uses some Chrome extension to get ChatGPT to generate illustrations for each leacture done in parallel in separate Chrome tabs allowing the author to watch while it was being generated. Cowork also watches and provides comments on the stage reached in the image generation task.
  • This needs a paid subscription and probably a higher level paid subscription. Perhaps even ChatGPT parallel image generation would need paid subscription but I am not sure about that.
  • It was very impressive to see the ability to have Cowork (from a desktop Claude app, IFIRC) pick up all readable data (not binaries) in a folder and understand it.
    • For my free tier usage of ChatGPT and Gemini, I have to individually upload either the files or the files contents (latter for ChatGPT as uploading a file very quickly exhausts a particular chat forcing a new chat to be created with loss of context built in previous chat). That is a time consuming task.
    • I have not tried out Gemini CLI or ChatGPT Codex in the past six months or so. In the past, perhaps over six months back, I had tried GitHub CoPilot within VSCode to study and modify the project I had loaded in VSCode. But I ran out of the free tier quota for that very quickly and had to switch to AI chatbot. Then I decided to stick to AI chatbots as even when they exhaust some free tier for a current model, they continue the same chat with an older model. After switch to older model, I have not yet encountered a limit with both ChatGPT and Gemini even when I have used it for long on some days.
    • I need to try out Gemini CLI and ChatGPT Codex now, which, I believe, are intended to provide project-level understanding by ingesting files directly from a working directory, without having to individually upload files or their contents as in the chatbot case, and see whether I am able to get a fair amount of work done with them before encountering free-tier limits.
    • I have to mention that over six months back, I had some trials with Gemini providing an online project kind of structure for uploaded files. But a big issue I faced was that Gemini files would sometimes be out of sync with my project files. I also used Google AI studio (over six months back) but did not find it effective enough for some reason I don't recall now. So I decided to switch to Gemini and ChatGPT chatbots where I had to do some manual work of uploading content but I felt more in control of what was going on and developed a work style that fitted in with AI chatbot limitations.
    • Related ChatGPT chat extracts: Claude Code can ingest and reason over entire folders or repositories
    • Related ChatGPT chat extracts: Why I have been using ChatGPT and Gemini chatbots mainly for AI assisted work
Claude Code Tutorial for Beginners: Build App with AI (2026), ~22 mins by Mikey No Code, 25 Jan 2026. This is the same YT video mentioned earlier.
  • This is more of a step-by-step tutorial for a coding project.
  • Shows VSCode extension for Claude Code. Installing it brings in a chat panel inside VSCode like GitHub CoPilot.
  • At 3:56 in the video, the author enters a simple prompt to create a website like wikipedia using React and Node.
  • Claude then shows the various source code files it creates for the project. It creates a React frontend and an Express backend. It also creates sample articles on JavaScript, React, Node.js, Typescript and Python. It then provides the commands to be run to start the backend and frontend. It starts the backend on its own and leaves starting the frontend to the user. [Later in the video, we see that Claude can start the frontend too using an English language prompt like launch the app.]
  • He then asks Claude to add dark mode feature to the app which it does.
  • End to end testing of the app is not covered. Even Jest testing (snapshot testing) is not covered.
  • At 9:13 in the video, he provides the prompts to make a Math-focused homework helper React Native and Expo app which has only frontend without functionality.
  • Claude creates the app with lot of Math examples shown in the UI which it seems to have generated on its own.
  • At 10:17 in the video, he provides a prompt to add photos and provide preview of them.
  • At 14:20 in the video, he provides a prompt to integrate OpenAI in the app with OpenAI doing the scanning of the image, picking up the questions and answering them with a detailed breakdown. Hmm. I wonder why he did not use Claude after the questions were picked by OpenAI, to take the answering part itself.
  • He tests this feature with an example image. It is impressive to see how OpenAI analyzes the Math problem of complex numbers and provides a detailed step-wise solution, and how Claude provides the UI for this solution. Open AI API key usage comes into play which is not free. 
  • Then he gives a prompt to add functionality to save questions including associated images and their answers. Claude does the work.
  • At 17:42 in the video, he provides a prompt to add a learn section with a range of Mathematics courses for various levels. Claude generates such a section. Now it is not clear how good this generation is, as the prompt was very general. Would it fit within syllabus of some well known education board, I wonder? ... I had a chat with ChatGPT on this which I plan to put up as a separate post. Meanwhile, here's the summary it provided:
    • AI tools like Claude can already generate a substantial portion of syllabus-oriented learning content for an educational app, especially in structured subjects like Mathematics. While a vague prompt may produce only a demo-level, toy-like result, carefully constrained prompts targeting a specific board and grade can yield 80–90% usable draft material. This represents a major shift, because producing original, copyright-safe educational content traditionally consumes enormous time and effort. Subject-matter experts are still essential, but their role shifts from writing everything from scratch to reviewing, correcting, and refining AI-generated drafts—dramatically improving productivity without compromising standards. Related ChatGPT chat extracts: AI Generated Learn Content for K-12 Mathematics
  • At 18:58 in the video, he gives a prompt to add a quizzes feature to the application and Claude does it.
  • Then he does manual end-to-end testing of the app. Hmm. In the video, he tests the app in desktop Chrome by switching to the mobile device view via DevTools. Perhaps Playwright could be used to automate such end-to-end tests. I wonder why he did not ask Claude to create Playwright test scripts for it, execute those scripts and check the output logs. Perhaps that would have moved the focus of the tutorial video to more intricate release level work which the author may not have wanted to show.
How to run Claude Code for free, ~3 mins, Jan 2026, AI for Everyone. Covers how to use Ollama which has a free tier to run Claude Code with a free model downloaded on user's computer. The video appears to use an AI-generated voice narration. He gives the example of qwen3-coder model which has 19 GB download size but uses some other model finally (Deep seek IFIRC).
  • As per ChatGPT, in my case, running such large local models is not very practical. With my current PC configuration (Core i3 CPU, 16 GB RAM, and a 240 GB SSD), even a single ~19 GB model download is significant from a disk space perspective, and running it locally would likely be resource-constrained. For this reason, using locally hosted AI models via tools like Ollama does not seem well-suited for my setup.
  • As per ChatGPT, to comfortably run a large coding-focused LLM such as qwen3-coder locally using Ollama, a fairly capable system is required. Practically speaking, this means a modern multi-core CPU (at least mid-range i5 / Ryzen 5 class or better), 32 GB of RAM for smooth operation, and ample fast storage—preferably an SSD with at least 100–150 GB of free space to accommodate models, caches, and updates. While GPU acceleration is not strictly mandatory, a discrete GPU with sufficient VRAM can significantly improve responsiveness. Without such resources, running large local models tends to be slow, memory-constrained, or impractical for sustained development use.
  • As per ChatGPT, in practice, cloud-hosted AI models—including those accessed via platforms such as Ollama—are paid in almost all serious or sustained use cases. While many mainstream AI platforms such as ChatGPT, Gemini, Claude, and Copilot offer free tiers, these are primarily intended for interactive, human-driven use through their chat interfaces. API access, which is meant for programmatic use by applications, scripts, CI pipelines, or autonomous agents, typically falls outside the scope of free usage. This distinction explains why AI tools may appear free at the chat level, while real-world software integration and production workflows almost always incur costs.
  • Related ChatGPT chat extracts: Impractical for me: Using Claude code for free with Ollama and free model.

Comments

Post a Comment

Archive

Show more