pg-vector

U-PG-RAG, a campus Q&A backend system, is built on PostgreSQL (PG).U-PG-RAG是一个基于postgresql(PG)构建的校园问答后端系统

This project demonstrates how to implement a hybrid search engine for Retrieval-Augmented Generation (RAG) using Postgres with PgVector. It showcases the use of asynchronous streaming with Groq's function calling capabilities in a FastAPI application.

A starting point for building custom agentic LLM applications using Open Source tooling and models. Incorporates Ollama, Open WebUI, Langchain, Streamlit, Chroma, & PGVector using Docker and Docker Compose and optionally Codespaces.

LangChain abstractions backed by YugabyteDB Distributed SQL Backend

The modern web development landscape is plagued by a peculiar paradox: despite the abundance of UI components and design systems, developers still spend countless hours reimplementing similar interfaces. S0 addresses this challenge by introducing a novel approach that combines advanced vector search capabilities.

Agentic Graph RAG system built on Pydantic AI combining vector search (pgvector/Supabase) with knowledge graph (Neo4j/Graphiti). Features intelligent query routing, multi-hop reasoning, long-term memory, and tools for web search, SQL, code execution, and image analysis.

AI-powered job search platform with real-time preference filtering.

Explainable semantic search for real estate marketplaces, combining vector search, cross-encoder reranking, and SHAP/LIME explanations to deliver relevant and transparent property results.

FastAPI + AutoGen 7-agent sports chatbot with pgvector RAG, Tavily MCP search and SSE streaming.

DEMO using griptape package to build an agent to interact with model garden on GCP using public API.

Agentic RAG–powered assistant built with LangGraph that answers user queries from a knowledge base and guides users to create, update, or cancel Google Meet calls, managing availability and sending meeting links via email.

Enterprise AI Knowledge Platform powered by RAG, combining BM25 keyword search and dense vector retrieval with PostgreSQL, pgvector, and cross-encoder reranking to deliver more relevant, context-aware answers from enterprise documents.

Incremental RAG data pipeline: Airflow-orchestrated ingestion, chunking and embeddings into PostgreSQL pgvector (HNSW), re-embedding only changed chunks, with a retrieval-quality gate (recall@k, MRR).

Monocle is a multi-modal embedding service designed for easy integration into modern applications. It provides HTTP API endpoints for generating text and image embeddings using state-of-the-art models. Monocle is ideal for semantic search, recommendation, and AI-powered content understanding.