🎬 Titro App

Translation platform that lets users upload .srt files, translate them into multiple languages using OpenAI models, and download the results as a zip archive.

During a routine medical appointment, the practitioner mentioned needing subtitles for over 1 000 training videos, each in five languages. It was considered a months-long manual task. I left that day thinking: how long would it take to automate this?

Ten days later, Titro was live.

This project was built almost entirely through a structured, prompt-driven workflow — but not in a magical “AI builds the app” way. It was more like pairing with a tireless technical co-pilot.

The loop began with ChatGPT. Before writing any code, I used it to pressure-test the concept: asking architecture questions, exploring infrastructure tradeoffs, and clarifying feature boundaries. This back-and-forth helped me refine the scope, spot edge cases early, and converge on a solid plan.

Once confident in direction, I had ChatGPT generate .mdc files — technical, step-by-step prompts tailored for Cursor, my primary development environment. These weren’t vague user stories or acceptance criteria; they were precise instructions on how to implement backend services, database migrations, cron jobs, and React components in my stack. Each file mapped to one atomic feature or update, and I would feed it directly into Cursor to generate compliant, context-aware code.

For both backend and frontend, I also had ChatGPT produce a linear step-by-step roadmap. These roadmaps lived as reference .mdc files and guided my progress: I’d pick the next step, clarify it with ChatGPT if needed, and then prompt Cursor to implement it — keeping business logic clean and structure consistent.

Frontend work added another layer. Before coding, I asked ChatGPT to generate prompts for Vercel’s v0, which created UI components matching my needs. Sometimes I’d take the raw design output as-is; other times I used the screenshots and structure as references in my Cursor prompts. It was fast, iterative, and always aligned with the stack and component library I’d chosen.

The end result? A project built systematically, with each feature distilled down to a clear, testable unit of work — and a dev experience that felt more like orchestration than brute-force building.