BIJIO
BIJIO has been, and continues to be, an ongoing attempt to marry multimodal digital narrative mediums with generative AI. Specifically, we are trying to create infinitely expandable, immersive, non-railroaded, player-customised worlds.
This is a technically challenging project. LLMs cannot generate the entropy that makes media interesting. Projects like c.ai and AI Dungeon get around this by having the player directly interface with the engine through language. The human plays more of a role in shaping their world than a machine. But this is not immersive in the way a visual novel might be.
Moreover, LLMs are much worse at maintaining a coherent world over time. Characters need persistent memories, locations need stable properties, actions need consequences, and the model needs to distinguish between what has actually happened and what would merely make a plausible continuation.
BIJIO therefore treats the language model as one part of a larger narrative system. Structured world state, character histories, player decisions and generated media sit outside the model and are selectively fed back into it as needed. The broader challenge is balancing freedom with consistency. Too much structure recreates the railroaded games we are trying to move beyond, while too little produces a world that quickly contradicts itself. BIJIO is an experiment in finding the point between those two extremes.
TUFF.BET
TUFF.bet is a social screen-time app built around competition, because self-control does not seem to work.
Friends create leagues, choose which apps to restrict, and buy back minutes of screen time as they use them. That money enters a shared pot, and whoever uses the least paid screen time wins it.
The idea is that conventional screen-time tools rely on awareness and willpower, which are weak incentives. TUFF instead creates continuous monetary friction, social accountability and competition. Every extra minute has a cost, and every unnecessary unlock makes it harder to beat your friends.
QualTA
QualTA is an internal project I built for a professor of mine and for Princeton qualitative researchers who wanted an AI-first approach to language coding.
The platform uses a structured coding framework where each code includes definition, inclusion, and exclusion criteria. It integrates OpenAI's API for pattern identification and thematic analysis while maintaining transparency in the coding process.
