Chandra: an AI-assisted Panchang PWA from idea to production

Starting point

The project was an AI capability experiment inside a culturally complex domain.

The core objective was to see how capable AI could be when used across research, product strategy, interface design, accessibility review, coding, testing and launch preparation. Panchang and moon tracking made the experiment demanding because the domain combines astronomy, Hindu calendar conventions, regional practices, Sanskrit terminology and everyday usability.

The design challenge was not simply to show more Panchang data. It was to make accurate cultural information easier to understand for people who either use printed Panchang daily or are curious but intimidated by existing digital tools.

Claude as research partner

Claude helped map the opportunity, competitors and platform strategy.

The first prompt explored whether a responsive web application could help people track the moon with Indian cultural context, what could be built, what already existed and whether the product should begin as a web app or cross-platform mobile app. Claude's response created a starting landscape and helped frame the PWA-first direction.

Human grounding

Informal discovery and manual testing grounded the AI-assisted process.

Before development, conversations with 4 to 5 people explored how they use Panchang and what they do with the data. These were not formal research interviews, but the discussions were detailed enough to start the design process. After the PWA was built, 15 people tested it: 3 daily Panchang users and 12 people who were interested but did not know how to use Panchang confidently.

Design strategy

Chandra was shaped as a basic, understandable Panchang instead of a complete expert almanac.

The first version needed to be useful without overwhelming people. The experience was separated into four clear areas: Day View for the daily ritual moment, Calendar for planning, Panchang for detailed information and Settings for city, calendar system and notification control.

Before building the app, I used Codex connected to Figma through MCP as a brainstorming partner. Based on the research direction, Panchang user needs and product constraints, I prompted Codex to generate the right low-fidelity screen set and detailed user flows in Figma. The screens helped shape the daily moon moment, calendar scan, Panchang detail view, settings controls and shareable daily message, while the user flows clarified the navigation model, decision points and edge cases before the PWA was developed.

Design consistency and accessibility

Design-system and accessibility work clarified where AI output had drifted.

Using the low-fidelity screens and user flows as the starting point, I began developing the application. The first version of the developed screens looked consistent at a glance, but detailed review showed different colors, styles and icon choices across the app. A design system defined colors, typography, spacing, cards, toggles and iconography. Lucide icons replaced most UI emojis, while moon phase emojis were retained where they carried semantic meaning. The first version was also tested for accessibility, which helped identify usability gaps before launch readiness.

AI helped produce and critique the product, but human review was still needed to notice inconsistency, define the design-system work and request accessibility testing before treating the app as launch-ready.

Codex as implementation partner

Codex helped turn the research direction into a working, tested PWA.

The app required more than screen generation. Codex supported the build across React screens, calculation logic, service worker behavior, Firebase notifications, accessibility remediation, build verification and Google Play TWA preparation.

Feedback-driven roadmap

Testing helped separate immediate improvements from future platform bets.

Not every story point belongs in version 1. The testing feedback helped identify what could be incorporated quickly, what required deeper calculation validation and what depended on a future native application.

Google Play closed testing

The PWA moved into closed testing through a Trusted Web Activity wrapper.

Chandra is currently in Google Play closed testing. Most product and design updates can continue through the PWA deployment flow, while wrapper-level issues such as launcher icons, splash screens and Android signing require Play Store-specific work.

What the project showed

AI accelerated the full product cycle, but judgment and user grounding still mattered.

The biggest learning was not that AI can simply build an app. It was that AI can accelerate research, planning, critique, implementation and release preparation when the designer actively directs the work, checks the output and validates the product with real people. Chandra became a working test of AI as a capability multiplier, not a replacement for product judgment.

For culturally sensitive products, speed is useful only when paired with trust. AI helped the work move fast, but the product became credible only when Panchang users, interested beginners, accessibility review and Play Store testers shaped the next decisions.