Sojourn AI — Cinematic Travel Memory Platform

Sojourn AI transforms a travel dream or emotional feeling into a full cinematic campaign — persistent, evolving, and beautiful.

Describe your journey in one sentence. Sojourn builds a complete visual universe: a cinematic poster, 4 documentary-style daily itinerary frames, 3 polaroid memory snapshots, and a video trailer — all wired together as a connected Melius workflow.

What makes it different: the canvas doesn't reset. Every follow-up prompt evolves the SAME cinematic universe using node_update instead of regenerating from scratch. "Make it rainier." "More hopeful ending." The world deepens over time.

--- How I built it --- • Built a FastAPI backend that orchestrates the full Melius workflow via MCP • Used 7 node types: custom_text (source prompt), text (director/outline), image ×5 (poster + itinerary days + polaroids), video (trailer) • Connected all nodes with edge_create / bulk_create_edges to form a proper directed graph — source → director → parallel branches → trailer • Used canvas_plan_layout before node creation so the canvas renders as a clean visual workflow • Wired Gemini as the creative director layer to structure prompts into cinematic directions before Melius generation runs • MongoDB persists conversationId ↔ canvasId ↔ projectId so every session continues the same canvas • Deployed full-stack with a Next.js cinematic frontend

--- Feedback on Melius --- The MCP server is genuinely powerful for agentic workflows — bulk_create_edges and canvas_plan_layout are underrated tools that made the parallel graph structure possible. The ability to update nodes instead of regenerating is what unlocked the "persistent universe" concept entirely.

Project link: https://sojourn-melius.vercel.app Screen recording and Canvas screenshot