Viberary is a side project created by the author to find books based on “vibe”. The project uses a semantic search engine and a two-tower semantic retrieval model to encode user search queries and book corpus. The author explains the process of building Viberary, mentioning the challenges and decisions made along the way. The architecture involves data collection, modeling, generating embeddings, indexing, and model inference. The author emphasizes the importance of thoughtful engineering and design choices in building a machine learning application. The project’s tech stack primarily consists of Python, DuckDB, SBERT, Redis, Flask, and Bootstrap. Overall, the author highlights the learning experiences gained from the project and the satisfaction of shipping a product they built.
https://vickiboykis.com/2024/01/05/retro-on-viberary/