rag-service

Load per-tenant FAISS indexes and retrieve documents for Hungarian voice assistant queries.

Updated May 20, 2025

Search Tags:#rag #LLM #FAISS #vector-index #Hungarian #voice-assistant

Ready to run

npx skills add https://github.com/papdawin/customer-service-assistant --skill rag-service

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: rag-service
Source: https://github.com/papdawin/customer-service-assistant/tree/main/rag
Command: npx skills add https://github.com/papdawin/customer-service-assistant --skill rag-service

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Enable tenant-aware retrieval-augmented generation for a Hungarian voice assistant, delivering accurate, context-driven answers from company documents.

Core Features & Use Cases

Contextual Retrieval: Loads a per-tenant FAISS index and retrieves relevant documents to answer questions in real time.
Context Formatting: Formats retrieved snippets into a compact context window for the LLM.
Use Case: A customer asks for store hours for Tenant X; the system returns precise hours and contact information from tenant documents.

Quick Start

Start the rag-service, load the tenant index, and query with a sample prompt like "What are the opening hours for Tenant X today?"

Frequently Asked Questions about rag-service

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I build a tenant-aware RAG system for a voice assistant?▼

To build tenant-aware retrieval-augmented generation, you load a per-tenant FAISS vector index and retrieve contextual documents to answer company-specific questions in real time. The system formats retrieved snippets and invokes a shared LLM prompt with timing telemetry.

How does FAISS work with retrieval-augmented generation for multiple tenants?▼

FAISS enables tenant-aware retrieval by loading a dedicated vector index per tenant. When a query arrives, the system retrieves relevant documents from that specific tenant's index, formats the context window, and passes it to a shared LLM for accurate, company-specific answers.

Can I use a shared LLM prompt across different tenants in a RAG pipeline?▼

Yes, this RAG pipeline uses a shared LLM prompt across all tenants while maintaining isolation through per-tenant FAISS indexes. Context formatting ensures each tenant only retrieves and submits their own company documents to the shared LLM prompt.

What is the best way to provide real-time company knowledge from documents to users?▼

Real-time company knowledge retrieval uses a FAISS vector index to find relevant document snippets based on user queries. The system formats these snippets into a compact context window for the LLM, returning precise answers like store hours or contact information instantly.

Does retrieval-augmented generation support Hungarian language voice assistants?▼

Yes, this retrieval-augmented generation system is specifically designed for a Hungarian voice assistant. It processes spoken company-specific questions in Hungarian, retrieves matching contextual documents, and generates accurate real-time responses using the shared LLM prompt.

What are the limitations of using a per-tenant FAISS index for RAG?▼

Using a per-tenant FAISS index requires loading and managing a separate vector index for each tenant, which increases memory usage and operational overhead. The system is optimized for instant company knowledge retrieval but depends on the quality of tenant-specific indexed documents.