The Workstation

Total hardware investment of approximately €5,000 — ensuring data sovereignty, localised control, and independence from commercial cloud services.

Processor
AMD Ryzen 9 9900X
GPU
NVIDIA RTX 4090 · 24GB VRAM
Memory
64 GB DDR5 RAM
Storage
1 TB NVMe SSD
OS
Windows 11 Pro
Total Investment
~€5,000

RAG Pipeline

The Retrieval-Augmented Generation pipeline actively queries a curated document repository at inference time, constraining outputs within expert-verified knowledge boundaries.

REPOSITORY 71 PDF documents Blue Skills (CK) + ECEC (PK) 1000-char chunks EMBEDDINGS llama3.2:3b via Ollama VECTOR DB ChromaDB k=5 similarity search GENERATION gpt-oss:20b MoE · MXFP4 quantised Apache 2.0 · via Ollama 3-retry mechanism OUTPUT Lesson plans · Curricula Stories · Songs · Activities USER QUERY Educator prompt SYSTEM PROMPT v10 · LOF embedded

Platform Components

Backend

Framework
Python · Flask
Orchestration
LangChain · LangGraph
Persistence
SQLite

Frontend

UI Framework
Vue.js
Features
Chat · Admin · Multi-thread

Why Local Hosting?

The choice of a locally hosted, open-weight model ensures data sovereignty, provides explicit control over the knowledge boundary, enables transparency of the retrieval process for expert auditing, and eliminates dependency on commercial cloud services — consistent with EU Ethics Guidelines for Trustworthy AI.