Search, from scratch.
An information retrieval engine built ground-up: BSBI & SPIMI indexing, compressed postings, BM25 with WAND pruning and LSI semantic search.
Four rankers, one query.
Same index, same query: see where TF-IDF, BM25, WAND and the adaptive lexical + semantic hybrid agree, and where they diverge. Hover a document to trace it across columns.
Inside the inverted index.
Nine index builds, every combination of indexing algorithm, term dictionary and postings codec, each stored in its own directory. Inspect how a postings list is laid out bit-by-bit, and rebuild any index live.
Postings compression
Total size of the postings file for the whole collection.
Postings inspector
Look up a term to see its postings list, the gaps between doc IDs, and how each codec writes them.
Build matrix
Indexer × dictionary × codec. Each cell is a real index on disk.
Measured, not guessed.
30 benchmark queries scored against human relevance judgments (qrels), ranked to depth 1000. Metrics: rank-biased precision, discounted cumulative gain, normalized DCG and average precision.