PSE - Pdf Semantic Search

This personal semantic search engine allows you to upload PDFs, markdown files, and notes and then search them by meaning instead of exact keywords. It does this by chunking your documents, generating vector embeddings with a transformer model, storing these in Chroma, and exposing FastAPI endpoints that a React frontend calls for upload and search. When you query, the backend embeds your question, finds the most similar chunks in the vector store, and re-ranks them using recency and metadata before returning results. Building it required integrating multiple pieces: file handling, text extraction, embeddings, a vector database, an HTTP API, and a modern frontend UI. In this project, you learned how vector-based semantic search pipelines are designed from end to end, how small details in chunking and embeddings affect relevance, and how to deploy a full-stack AI-powered app.