Project Madras
AI-Driven Historical World Reconstruction for VR
What It Is
An AI system that reconstructs historical Madras (Chennai) across time (1800–2000) as explorable VR environments, combining archival data, structured world modeling, and deterministic multimodal generation.
Immersive VR worlds break when they hallucinate. This project explores how LLMs can be constrained into reliable world builders — producing historically grounded environments suitable for education, cultural preservation, and social exploration.
What I Built
- Designed EvidencePack → SceneSpec → PromptPack schemas to separate facts, interpretation, and visual intent
- Built a RAG-to-Scene compiler that converts archival sources into VR-ready scene specifications
- Implemented deterministic image and short-video generation (multi-view, spatially consistent)
- Designed foundations for NPC agents that speak and behave according to time-period-specific social context
- Structured the system to support community-sourced memory layers without corrupting canonical history
Tech Stack
PythonLLMsRAGStable DiffusionJSON SchemasVR Pipelines
Working product demo showing how AI can safely generate immersive historical spaces. Applicable to cultural worlds, educational VR, and long-form social environments.