🥤 RAGLite is a Python toolkit for Retrieval-Augmented Generation (RAG) with DuckDB or PostgreSQL

PyLate efficient inference engine

High-quality search for AI-native applications.

Official repository of TACHIOM.

[ICML'26] LEMUR reduces multi-vector retrieval for late interaction models such as ColBERT into regular single-vector retrieval.

Fused Triton kernels for late-interaction (MaxSim) scoring

High-performance late-interaction retrieval engine for on-prem AI. ColBERT/ColPali multi-vector search with Rust fused MaxSim, Triton GPU kernels, ROQ quantization, LEMUR routing, WAL-backed CRUD, and a FastAPI server — single machine, CPU or GPU.

AccInt - local-first MCP Work Model for coding agents that learns from real outcomes.

Python library for MUVERA multi-vector retrieval via Fixed Dimensional Encodings. ColBERT / ColQwen2 / ColQwen3.5 compatible.

A physics-grounded, agent-driven digital twin for HP Metal Jet S100 3D printer

ColBERT inference in C++ (ONNX Runtime + tokenizers-cpp) without PyTorch.

Serve Jina Embeddings v4 multi-vector (ColBERT-style) multimodal text+image embeddings from a stock vLLM OpenAI server via an out-of-tree plugin, with configurable image fidelity.

ColFastVLM: Towards low-latency indexing in visual document retrieval

MCP server exposing late-interaction (ColBERT-style MaxSim) document retrieval over a local folder — pluggable encoder, runs offline by default.

Multi-modal Contextual Late Interaction RAG with Triple Filtering (Multi-modal CoLRAG-TF) retrieval strategies on Japanese Natural Disaster documents within multiple modes included in Text, Figure, Table, Photos, Map.

Repo for portfolio, containing working redirects to all projects.

ColPali-style multimodal late-interaction retrieval (text → document-image patches via MaxSim) — numpy reference implementation

ColBERT-style late-interaction MaxSim scoring for multi-vector retrieval — tiny, zero-dependency Rust crate