Case study

ZenithAI

An AI product generating personalised development reports from NASA JPL astronomical data. An interactive React configurator, LLM-driven content generation, encrypted online payments and full PL/EN bilingualism.

ZenithAI — Web app

At a glance

Year
2025
Scope
AI product, configurator, payments, two languages
My role
Product, full-stack, LLM integration
Stack
Next.jsReactAI / LLMStripePostgreSQLi18n

Case study

01The problem

The report comes from pairing hard NASA JPL astronomical data with text generated by a language model. A model will happily write a graceful sentence about something the data never contained — and the customer pays up front and receives a file nobody will correct afterwards. The work was in separating what the model is allowed to invent from what has to come out of a calculation.

02What I built

  • Every numeric value is computed application-side from the JPL data; the model receives them as settled facts, not as a sum to work out.
  • The model is responsible for the narrative layer alone — the shape and language of the report, never the inputs.
  • A React configurator walks the user through step by step and validates the inputs before payment is allowed.
  • Full PL/EN bilingualism at the level of content and of the generated report, not just the interface.

03What it does now

  • The figures in the report are reproducible — the same input set always yields the same result.
  • A purchase ends with the file delivered, with nobody on the seller's side involved.
  • Both language versions come out of a single generation pass.

04Why this stack

The line between the calculation and the language model is the most important architectural decision here. Without that line a model will eventually write a number nobody computed — and in a paid product that is not a risk you get to accept.

All projects
Let’s start

Got a project in mind?

Describe it in a few sentences — I reply within 24 hours with a free quote and a proposed stack.