The developer launched postvid.ru, a service designed to automate video transcription, summarization, and content creation. The platform processes video links or files to generate summaries, tables, presentations, and social media carousel cards. The backend is built on FastAPI, utilizing SQLite for managing user accounts, subscriptions, and payment data.
The technical pipeline relies on faster-whisper (small model) for speech-to-text conversion on CPU-based servers. For summarization, the system uses YandexGPT or OpenAI-compatible APIs. To ensure high availability, the developer implemented a fallback mechanism: if the cloud LLM fails, the system switches to a local model via Ollama or a heuristic-based offline algorithm.
The infrastructure includes Docker Compose and Caddy for deployment and automated TLS certificate management. Payments are processed through the ЮKassa service. The architecture handles tasks in a thread pool, with job statuses tracked in process memory and retrieved via polling through the API endpoint.
The developer addressed security concerns by implementing SSRF protection to validate public URLs. The system strictly filters schemes to http or https and restricts access to internal network resources. Users are notified on the results page about which engine—cloud, local, or offline—was utilized to process their specific request.
