RAG Service

Retrieval-Augmented Generation service for the TFG educational chatbot. Provides document management, semantic search, and vector storage using Qdrant and Ollama embeddings.

Overview

graph LR
    subgraph "RAG Service :8081"
        API[FastAPI API]
        FileLoader[File Loader]
        DocProcessor[Document Processor]
        EmbeddingService[Embedding Service]
        VectorStore[Vector Store]
    end
    
    subgraph "External Services"
        Ollama[Ollama :11434]
        Qdrant[(Qdrant :6333)]
    end
    
    API --> FileLoader
    FileLoader --> DocProcessor
    DocProcessor --> EmbeddingService
    EmbeddingService --> Ollama
    EmbeddingService --> VectorStore
    VectorStore --> Qdrant

Key Features

• 📄 Document Management: Upload, list, and organize documents by subject and type
• 🔍 Semantic Search: Find relevant content using natural language queries
• 🧩 Automatic Chunking: Split documents for optimal retrieval
• 🔢 Embedding Generation: Convert text to vectors via Ollama (nomic-embed-text)
• 🗃️ Vector Storage: Efficient similarity search with Qdrant
• 📊 Prometheus Metrics: Built-in monitoring and instrumentation

Quick Start

Local Development

# Install dependencies
cd rag_service
pip install -e ".[dev]"

# Set environment variables
export QDRANT_HOST=localhost
export OLLAMA_HOST=localhost

# Start the service
uvicorn rag_service.api:app --reload --port 8081

Docker

# Start with dependencies
docker compose up -d qdrant ollama rag_service

# Initialize embedding model
docker exec ollama ollama pull nomic-embed-text

API Access

• API: http://localhost:8081
• Health: http://localhost:8081/health
• Metrics: http://localhost:8081/metrics

API Endpoints

Method	Endpoint	Description
GET	/health	Health check with Qdrant status
POST	/search	Semantic search with filters
POST	/index	Index documents
GET	/files	List available files
POST	/upload	Upload and index file
POST	/load-file	Index existing file
GET	/subjects	List subjects
GET	/subjects/{asignatura}/types	List document types
GET	/collection/info	Qdrant collection stats

Document Organization

documents/
├── logica-difusa/
│   ├── apuntes/
│   │   └── tema1.pdf
│   └── ejercicios/
│       └── practica1.md
├── iv/
│   ├── teoria/
│   │   └── docker.pdf
│   └── examenes/
│       └── examen-2024.pdf

Configuration

Variable	Default	Description
QDRANT_HOST	qdrant	Qdrant server hostname
QDRANT_PORT	6333	Qdrant server port
OLLAMA_HOST	ollama	Ollama server hostname
OLLAMA_PORT	11434	Ollama API port
OLLAMA_MODEL	nomic-embed-text	Embedding model
EMBEDDING_DIMENSION	768	Vector dimension
TOP_K_RESULTS	5	Default search results
SIMILARITY_THRESHOLD	0.5	Minimum similarity score
CHUNK_SIZE	1000	Document chunk size
CHUNK_OVERLAP	200	Chunk overlap

Dependencies

• Qdrant: Vector database for similarity search
• Ollama: Local LLM inference for embeddings
• LangChain: Document processing and chunking

Documentation

Document	Description
Architecture	System design and data flow
API Endpoints	Complete API reference
Embeddings	Embedding service and models
Vector Store	Qdrant integration
Document Processing	Chunking and loading
Configuration	Environment variables
Development	Local setup and testing
Deployment	Docker and production

---

← Back to Services