Open-source inference server and production cluster for all the models your agent needs.

SPLADE: sparse neural search (SIGIR21, SIGIR22)

Neural Search

An easy-to-use python toolkit for flexibly adapting various neural ranking models to target domain.

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

Local corpus retrieval over your own text: BM25 + SPLADE v3 + ColBERTv2, with synthetic eval and an optional relevance judge

Optimized RAG Retrieval with Indexing, Quantization, Hybrid Search and Caching

Local-first RAG system with hybrid (dense + BM25) and SPLADE retrieval, hierarchical conversational memory, and document-aware reasoning.

Advanced RAG pipeline with hybrid dense+sparse search (fixed-size, recursive, semantic chunking), cross-encoder reranking, and RAGAS-based evaluation across all config combinations. Streamlit dashboard included.

A production-grade Multi-Agent RAG (Retrieval-Augmented Generation) system designed for scalable, low-latency, and reliable AI-powered retrieval. Built with hybrid search, cross-encoder reranking, intelligent query decomposition, semantic caching, adaptive LLM routing, and ONNX-optimized inference using Qdrant, Groq, Gemini, and BGE embeddings.

GPU-accelerated semantic search for your docs and source code - hybrid dense + sparse RAG on a local Qdrant backend, served to Claude Code and other MCP clients. The search companion to vaultspec-core.

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

Training Data Generator for SPLADE Model Fine-tuning

Research: signal-processing approaches to semantic search, treating documents as continuous signals

Official code and camera-ready analyses for PFW Task 8 at SemEval-2026 Task 8 (MTRAGEval).

QLoRA fine-tuning pipeline: Llama 3.1 & Qwen 2.5 for structured extraction — SPLADE sparse embeddings, vLLM FP8 inference, end-to-end data pipeline

LLM embeddings examples

Memsplora - An in-memory SPLADE (SParse Lexical AnD Expansion) content server with FAISS integration