RAG-powered AI chatbot
Admin-managed chatbot with RAG (Qdrant) and real LLM tool-calling: Symfony / API Platform backend, Nuxt / Vue frontend, Ollama models running locally or in the cloud.
view projectSPEC SHEET
The context
An admin-managed chatbot designed to answer from its owner's own knowledge base rather than from the model's knowledge alone, and able to trigger real business workflows during the conversation. A personal project to build a complete AI architecture, from document indexing all the way to production.
What I built
A Symfony 8 / API Platform backend that orchestrates the chat: vector search on Qdrant (RAG), FAQ answers, LLM tool-calling and AI provider selection. A Nuxt 4 / Vue chatbot component consumes the API and streams the answers back. Models run on Ollama, locally or in the cloud.
- RAG: document processing and indexing, collections, vector search on Qdrant
- Real LLM tool-calling: the model triggers business workflows during the conversation
- AI provider selection (Ollama or OpenAI-compatible endpoint) with automatic fallback
- Multi-user /admin backoffice to manage the chatbot
- Docker Compose stack routed by domain through Traefik
- CI (lint, tests, PHPStan, audit) and automated deployment with GitHub Actions