A search engine built from first principles: given a query, it finds and ranks the most relevant documents in a collection, the core problem behind every search box.
The system rests on the vector space model. It preprocesses documents and queries with NLP techniques, converts them into weighted term vectors, and computes relevance as similarity between them, producing a ranked result list. On top of that, it clusters the document collection to group related content and includes a web crawler for gathering new documents to index.
Rather than trust that it “seems to work”, the evaluation follows IR research practice: standard test collections with known relevance judgments, objective retrieval metrics, and several corpora.
Written in Python with a Streamlit interface for running queries and exploring results, and a full written report on the design and evaluation.