Case study

VisionMap

3D Earth interface that generates narrated video tours from prompts using an AI media pipeline.

Jan 2025 – Mar 2025
GenAI3DProduct

Overview

VisionMap combines an interactive 3D globe with an AI pipeline that produces location-based video tours. Users can select a region and generate a cohesive storyline and visuals.

Problem

Creating engaging, location-based video content takes time and specialized tools.

Solution

I integrated Unity (3D experience) with Python services and AI APIs to generate scripts, scene plans, and rendered video outputs.

Architecture

  • Unity front-end selects location + prompt
  • Python orchestrator calls LLM for script + structure
  • RunwayML generates/edits visuals → stitches final video

Tech stack

Unity for 3D interactionPython orchestrationOpenAI API + RunwayML

Key engineering decisions

  • Pipeline orchestration to keep creative steps deterministic and debuggable.
  • Separation of concerns: Unity UX vs backend generation.

Results

  • Render Time: 30s

Links

What I’d improve next

  • Add caching for reusable assets and rerun-at-step to reduce iteration time.
  • Add safety filtering and content constraints for production readiness.