sparse-retrieval

SeekStorm: vector & lexical search - in-process library & multi-tenancy server, in Rust.

A Python Search Engine for Humans 🥸

Unified Learned Sparse Retrieval Framework

🚀 Engram-PEFT: An unofficial implementation of DeepSeek Engram. Inject high-capacity conditional memory into LLMs via sparse retrieval PEFT without increasing inference FLOPs / DeepSeek Engram 架构的非官方实现。通过参数高效微调 (PEFT) 为大语言模型注入超大规模条件记忆，支持稀疏更新且不增加推理开销。

SPRINT Toolkit helps you evaluate diverse neural sparse models easily using a single click on any IR dataset.

Fast search index for SPLADE sparse retrieval models implemented in Python using Numpy and Numba

Lite weight wrapper for the independent implementation of SPLADE++ models for search & retrieval pipelines. Models and Library created by Prithivi Da, For PRs and Collaboration checkout the readme.

Provides a minimal PyTorch implementation of SPLADE

Optimised BAAI/bge-m3 serving with dense + sparse + ColBERT embeddings, async dynamic batching and pipeline GPU inference

Semantic search in one SQLite file. No model, no server at query time.

SPLADE (Sparse Lexical AnD Expansion) model fine-tuned for Portuguese text retrieval. Based on BERTimbau and trained on Portuguese question-answering datasets.

Python code to train SPLADE sparse retrieval models based on BERT-Tiny (4M) and BERT-Mini (11M) by distilling a Cross-Encoder on the MSMARCO dataset

AI-native long-term memory for Hermes Agent with AES-256-GCM encryption at rest, dense HNSW ANN, sparse and exact-text recall, metadata/range filters, and MinHash dedup, backed by an embedded or multi-client server columnar SQL database with 35 language bindings.

Demo for SenTrEv python package

Benchmarked RAG pipeline over 150 arXiv ML papers — compares BGE, MiniLM, MPNet, and BM25 across chunk sizes with MRR, Faithfulness, and Answer Relevance metrics. Powered by FAISS and Gemini.

OctoVector AI is a high-performance RAG system that combines dense and sparse retrieval with fusion and cross-encoder reranking to deliver precise, context-aware answers.

a grounded, constraint-aware conversational retrieval engine built for the SHL Product Catalog.

A controlled experiment evaluating whether hybrid (dense + sparse) retrieval surfaces evidence that dense-only RAG systems misrank—without changing generation behavior.

RAG routing research prototype: Feature-based classifiers learn when to use cheap vs full retrieval/LLM paths based on retrieval difficulty

Enterprise RAG (Retrieval-Augmented Generation) system featuring document chunking, hybrid search (BM25 + Dense Embeddings), vector search, reranking, tenant isolation, and RBAC. Built with Go and Python.