Viloi

Turn a writing tool into a working product

Viloi rewrites machine-generated text so it reads more naturally. Building it meant thinking beyond the prompt: how to respond when a model gives an unsuitable result, how people access a paid plan, and how subscriptions keep working after the first payment.

The decisions behind the product

Give the pipeline another way to try. I connected GPT-4, Claude and Gemini with fallback routing. When an output is flagged, the pipeline can retry through another model. That gives the product another path when one model's response is unsuitable.

Build for the whole subscription lifecycle. I used Stripe webhooks to handle plan changes, cancellations and failed-payment follow-up. Those events need to work just as reliably as checkout if people are going to pay for the product over time.

Carry the work through the interface. I designed and built the UI alongside the API, authentication and billing. It gave Viloi a visual identity of its own and let me take responsibility for the complete product.

My contribution

I built Viloi solo, from the interface and multi-model pipeline to JWT authentication and subscription handling. The engineering decisions covered both the writing feature and the ongoing work of running a paid service.

What that made possible

Viloi is live with paying customers. Model fallbacks provide a way to retry unsuitable results, and automated subscription handling keeps routine billing changes from becoming manual work.