# learning-rag-agent

**Repository Path**: samhula/learning-rag-agent

## Basic Information

- **Project Name**: learning-rag-agent
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-06-28
- **Last Updated**: 2026-07-11

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Lifelong Learning RAG Agent

A Retrieval-Augmented Generation (RAG) system designed for continuous, lifelong learning. It ingests documents, stores knowledge in a graph database (Neo4j) and vector store (Qdrant), and answers queries using LLM-based synthesis with feedback-driven improvement over time.

## Architecture

The system follows a pipeline architecture. Documents are parsed by **MinerU**, split into chunks, and embedded via **OpenAI** embeddings. Both the chunks and their embeddings are stored in **Qdrant** (vector store) while entities and their relationships are stored in **Neo4j** (graph database). This dual-storage approach enables both semantic similarity search and structured knowledge graph traversal.

When a query arrives via the **Express** HTTP server, the system retrieves relevant context from both stores, synthesizes an answer using a configurable LLM, and logs the interaction. Users can submit feedback on answers, which is persisted for future fine-tuning. A web **Dashboard** provides a visual interface for managing the knowledge base, viewing config, and monitoring health.

## Quick Start

### Prerequisites

- **Node.js 24+**
- **Docker** (required for MinerU and Qdrant services)
- A running Neo4j instance (local or Docker)

### Installation

```bash
git clone <repo-url>
cd lifelong-learning-rag-agent
npm install
cp .env.example .env
# Edit .env with your API keys and service endpoints
```

### Running

```bash
# Command-line interface
npm run cli

# Web dashboard
npm run dashboard
```

## Configuration

Configuration is managed through environment variables in `.env` (see `.env.example`). Additionally, runtime configuration can be modified via the **Dashboard Config tab**, which exposes the same settings through a UI and persists changes to a local config store.

## API Reference

| Endpoint | Method | Description |
|---|---|---|
| `/api/query` | POST | Submit a query and receive a synthesized answer |
| `/api/feedback` | POST | Submit feedback on a previous answer |
| `/api/qa/create` | POST | Manually create a Q&A pair |
| `/api/qa/update` | POST | Update an existing Q&A pair |
| `/api/qa/delete` | POST | Delete a Q&A pair |
| `/api/qa` | GET | List all stored Q&A pairs |
| `/api/config` | POST | Update runtime configuration |
| `/api/health` | GET | Health check endpoint |

## Docker Deployment

```bash
docker-compose up
```

The `docker-compose.yml` file orchestrates all required services: the app server, Neo4j, Qdrant, and MinerU.

## License

MIT