The problem
Decrypta AI was designed to help creators and teams work backward from content they already have. A user could submit a text document, image, audio clip, or video and receive structured prompt material intended to capture its subject, style, or composition. The aim was to make existing media a starting point for a new AI workflow, rather than asking users to write a prompt from scratch.
What I built
The backend routes each format through its own preparation path before prompt generation. Repository code shows PDF text extraction, image description, audio transcription, and FFmpeg-supported video-frame analysis combined with transcription. The prototype used Claude, Gemini, and OpenAI models to generate and evaluate prompt candidates, and included a RAG-oriented graph/chat exploration around the extracted material. Redis-backed background tasks support longer processing jobs, while the Next.js frontend and Supabase services provide the surrounding account, file, and result experience. Propelius's case study also describes a dashboard, processing status, and Stripe subscriptions; those are product-scope descriptions, not verified production outcomes.
- Each media format has its own extraction path before prompt generation, with background jobs for longer processing. The project remained a prototype and was not launched to production.
