Project Description

An immersive, D&D-style adventure game built in 48 hours where the player becomes the hero. The application uses multi-modal AI to transform a user's selfie into a theme-appropriate, photorealistic character, and then dynamically generates a branching, cinematic story with unique scenes and narrated audio for every choice the player makes.

Responsibilities

• Architected and built a complete, end-to-end multi-modal AI application in a 48-hour hackathon, integrating text, image, and audio generation into a seamless user experience.
• Engineered a complex generative pipeline, using a text model (Gemini 1.5 Flash) to create a story structure that served as the input for an image model (Gemini 2.5 Flash) and an audio model (ElevenLabs).
• Developed a robust Python backend using FastAPI, implementing high-performance techniques like asyncio to run parallel image and audio generation tasks, significantly reducing API response times.
• Solved nuanced prompt engineering challenges, refining instructions to handle AI safety filters, ensure visual consistency, and enforce a valid JSON schema for reliable data flow.
• Built a dynamic, responsive frontend with React and TypeScript, managing complex game state and asynchronous API communication.
• Successfully deployed the full-stack, multi-language application to a live production environment on Heroku, managing environment variables, buildpacks, and static file serving.

Related Links

• Live Demo
• GitHub Repository

Technology