pyserini

↕️ Intuitive axiomatic retrieval experimentation.

Toolkit for domain-specific information retrieval experimentation

kkbox 2023 歌曲推薦競賽分數最高的解決方案（1 / 172）

A state-of-the-art Information Retrieval system for TREC ROBUST04 achieving MAP 0.3309. Features a Novel 4-Way RRF Fusion architecture combining BM25, Neural Reranking (Cross-Encoders), and LLM-Augmented Query Expansion (Query2Doc) via LiteLLM.

Straightforward SERPs for Lucene indexes

Neural Language Processing (NLP) project (AY 2022/2023)

A multi-agent retrieval-augmented generation (RAG) system for legal question answering, implemented with DSPy. The project includes a structured evaluation and performance comparison against standard baseline methods.

Summarizing my findings on Query Expansion using Generative AI by evaluating and comparing the results of different query expansion methods.

Learning-to-Rank on MS MARCO Passages: candidate generation from prebuilt indexes and re-ranking for QA search

Project for CSIS371

Dense retrieval on the BEIR FiQA benchmark with Pyserini, sentence-transformer embeddings, and a FAISS HNSW index

A Claude Code plugin + offline-first retrieval engine that indexes your project's own docs and code, then compares six retrieval strategies — contextual BM25, LLM-enriched lexical, dense ANN, Lucene BM25, hybrid RRF, and tree-sitter AST chunking.

TREC ROBUST04 ad-hoc retrieval system achieving MAP 0.4044 via a Seeded LLM Tournament (GPT-4o-mini listwise judge) and a Hybrid BM25+RM3+MonoT5+RRF pipeline. MSc final project, Reichman University.