MVP Launch - Hey Celestia

LGSE Limited partnered with Xenvio to design and build Hey Celestia as a production-ready MVP from the ground up, architected for real users, sensitive data, and scalability.

Architecture for Trust and Scale

We chose a serverless, event-driven architecture built on Firebase to minimise operational overhead whilst maintaining security and performance. This allowed Xenvio to focus on product-market fit rather than infrastructure management.

The mobile layer used React Native (Expo) for cross-platform delivery: one codebase for both iOS and Android without compromising native performance.

Security-first design

Rather than adding encryption as an afterthought, we architected end-to-end encryption from day one. Each conversation generates unique encryption keys held exclusively by the end user. Chat content never exists in plain text on our servers: not in logs, not in databases, not in transit. This was about earning user trust in a category where privacy matters.

AI as Infrastructure

We built the AI layer using LangChain and LangGraph to create modular, stateful workflows instead of monolithic prompt chains. This separation allowed us to:

• Insert moderation and guardrails at appropriate points in the flow
• Generate deterministic astrology data independently of AI generation
• Use calculated birth charts and astrological events as grounding context for AI responses
• Maintain explainability and consistency in outputs
• Enable debugging and troubleshooting capabilities for AI workflows

Rapid iteration

We built a custom admin interface for backend-managed prompt refinement. This architectural decision meant Xenvio could iterate AI tone, logic, and behaviour based on early user feedback without waiting days for app store review cycles. Version control and rollback capabilities provided safety during experimentation.

Revenue-Ready from Launch

Subscription infrastructure was integrated from the beginning. We implemented Apple App Store and Google Play subscriptions via RevenueCat with full compliance including account deletion flows and data portability requirements.

Building Hey Celestia reinforced that successful AI products need more than a chat interface - they need robust infrastructure. The difference between a prototype and a production AI product lies in moderation layers, deterministic grounding data, version control for prompts, and clear separation between AI logic and delivery mechanisms.

Using LangChain and LangGraph proved invaluable for building modular AI workflows that follow industry standards. This architectural choice gave Xenvio the ability to easily iterate and test system prompts and outputs during development and post-launch, without needing to understand the underlying implementation details. The separation of concerns meant AI behaviour could evolve independently from the mobile application.

This project validated that combining deterministic calculations with generative AI creates more reliable, explainable outputs than pure LLM responses. Users need both the flexibility of conversation and the trustworthiness of verifiable data.

One unexpected insight was how technical decisions became product differentiators. End-to-end encryption wasn't just a security feature - it became a trust signal that shaped how Xenvio could position the product in market. In categories handling sensitive personal data like journal entries and mood tracking, the ability to say ‘your conversations are encrypted and we can't read them’ changes the entire value proposition.

This taught us that architecture isn't just about how systems work internally - it's about what promises you can make to users. Building encryption from day one was easier than retrofitting it later, and it unlocked marketing angles that wouldn't have been credible otherwise.

For founders building products in trust-sensitive categories, early architectural choices don't just affect technical scalability - they affect how you can talk about your product and what competitive position you can claim.