I took over Storyling.ai, an AI language-learning app, while it was still a feature-rich prototype generated on a no-code/AI app-builder (Manus): running on managed hosting, a proprietary AI proxy and built-in storage, with none of the layers that make a product safe to sell and operate at scale. My mission was to turn it into a deployable, secure and maintainable product on independent infrastructure, then grow it end to end over three months. I opened with a full technical audit, delivered as a 53-page professional report that found 65 issues (31 critical or high) and framed the entire remediation roadmap.
The heart of the work was the leap from prototype to scalable production. I cut the cord with the proprietary platform and rebuilt every magic layer as direct, portable integrations: a single storage abstraction with three interchangeable backends (S3-compatible, Manus, local) and a central LLM client with automatic failover, timeouts and exponential-backoff retries. I containerized the app in a multi-stage Docker build deployed on Railway with health check and restart policy, closed the critical security holes (Helmet + CSP, locked-down CORS, tiered rate limiting, Stripe webhook signature verification, PBKDF2 auth, JWT sessions), and made the data layer hold under load by adding foreign keys and indexes across 71 tables where the prototype had almost none. The result is an architecture engineered to serve tens of thousands of learners without buckling.
On that foundation I built the technical centerpiece: a multi-provider AI generation engine that turns a theme, a target language, a CEFR level and a vocabulary list into either an audio podcast or a fully AI-produced film. The film pipeline chains structured story generation, multilingual voice-over with character-level timestamps, LLM character sheets, a deterministic seed and image-to-image identity anchoring for character consistency, then ffmpeg assembly with burned-in subtitles at 1080p, all kept resilient to provider outages and moderation through retry budgets, fallback cascades and circuit breakers. I then shipped a broad product around it: freemium and Stripe billing, spaced-repetition vocabulary, a context-aware AI chatbot, gamification, a B2B teacher and school space, and an offline-first PWA. Today Storyling.ai serves more than 35,000 learners and 2.5M+ generated stories across 19 languages, the proof that the production rebuild holds at scale.