Core Stack
- Next.js App Router for UI and API endpoints.
- PostgreSQL for persistent domain data.
- vLLM + Qwen 3.5 35B A3B FP8 for local inference.
- Docker Compose for reproducible local and VPS environments.
AI Storybook • Detailed Page
Technical Implementation
The platform is built as a production-like architecture by our student Daria Korneeva, combining modern web engineering with local LLM deployment.
Back to homepageData Layer
- Users: auth identity and role model.
- Books: title, plan JSON, generated text, publish state.
- Chat messages: multi-turn planning history.
- Generation jobs: status tracking, retries, and failure details.
Security Baseline
- Argon2id password hashing.
- JWT in httpOnly cookies with production Secure flag.
- Schema validation for all key request payloads.
- Rate limiting for auth, chat, and generation routes.
AI Pipeline
- Streaming planning chat for real-time UX feedback.
- Explicit finalization step to lock story plan.
- Structured plan validation before generation.
- Job-based long generation with controlled retries.
Deployment Model
- Local Docker first, VPS migration second.
- App as public entrypoint, DB and LLM on internal network.
- Single codebase and API contracts across environments.
- Designed for future monitoring and reliability extensions.