Adenstien
A shopping and advertising prototype pairing personalized product recommendations with AI-generated scripts, speech, and short video ads. Built for the Walmart Sparkathon.
From product information to media
The Adeinstien repository documents a Walmart Sparkathon shopping and advertising prototype. It describes monitoring interactions in a simulated store, recommending products, then creating personalized video ads from product information. The recommendation design names conservative Q-learning, Jaccard similarity, and cosine similarity.
The pipeline
The documented ad pipeline generates scripts, turns them into speech with text-to-speech, and assembles images into video. The repository lists React for the frontend and Flask with Python for backend and AI work. Its README links a project demo video.
Research connection
The supplied research record connects advertisement generation to Kolmogorov-Arnold Networks. The public project README does not establish how that work fits into this implementation, so the site does not infer a specific architecture or result.
What remains to document
The repository includes source and screenshots. The README describes possible distribution through YouTube, short-form video, and in-store screens; independently verified production deployment and measured outcomes are still unavailable.
Current status
Public project source and a demo video are available. Production deployment is not verified.
Recommendation · Conservative Q-learning · Jaccard similarity · LLMs · TTS · Video composition · Kolmogorov-Arnold Networks