Project case study
Interactive Three-Body reading platform
A participatory narrative system that lets readers inhabit four perspectives while a shared timeline, persistent memory, and live physics simulation keep the world coherent.
- Role
- Product design and full-stack development
- Stack
- Next.js, Three.js, LLM agents
- Interaction
- Four character perspectives
- Core constraint
- One coherent shared timeline
The problem
Branching stories often lose coherence as choices multiply. This project asks a different question: can a reader participate inside a complex narrative without breaking the causal structure that makes the world meaningful?
The system treats narrative continuity as an engineering constraint. Character knowledge, forced events, and physical state must remain aligned even as the reader changes perspective or rewinds a choice.
System design
- Four playable perspectives expose different knowledge and motivations.
- Cross-character memory preserves consequences beyond a single conversation.
- A controlled timeline cadence advances required world events.
- Context compaction keeps long sessions tractable while retaining consequential state.
- A live Newtonian three-body simulation gives the scientific premise a visible, inspectable form.
What I learned
The interesting work was not simply connecting an interface to a language model. It was designing boundaries: what the model may improvise, what the timeline must enforce, which memories should persist, and how the visual simulation can clarify rather than decorate the experience.
That same discipline matters in scientific and healthcare AI, where outputs need constraints, provenance, and interfaces that help people inspect what the system is doing.
Current status
A public branching demonstration is now available. Its character dialogue is pre-generated, so the experience makes no model calls and exposes no API credentials while preserving four ideological perspectives and multiple endings.