THE HOMELAB
A place to ask
“what happens if?”
Compact infrastructure. Useful experiments. A way to learn how AI, data, and operations fit together.
01 / THE BIG PICTURE
One lab. Different kinds of compute.
A NAS for applications and AI, Proxmox for virtualized workloads, and a compact RK1 cluster for MongoDB experiments. The RAG stack also uses managed MongoDB cloud services.
STORAGE & APPLICATIONS
NAS
HOSTED APPLICATIONS
Storage meets application hosting and local AI compute.
VIRTUALIZATION
Proxmox
VIRTUAL MACHINES
VMs provide separate environments for containerized workloads.
DATABASE EXPERIMENTS
RK1 cluster
A Turing Pi platform for repeatable builds and database measurements.
Atlas stores and searches the knowledge base. MongoDB’s AI Models API provides embeddings; managed reranking is an optional retrieval step.
Explore the RAG Gateway ↗02 / THE HYBRID AI LOOP
Local generation. Cloud-backed knowledge.
Hermes calls the RAG Gateway, which gathers evidence from managed MongoDB services and prepares the prompt for Ollama. The RTX GPU accelerates local model generation.
Agent tasks & chat requests
Ingestion, retrieval & evidence assembly
Local model execution
Compute for model inference
PUBLIC CLOUD / MONGODB MANAGED SERVICES
AI Models API
Embeddings for document passages and user questions.
MongoDB Atlas
Chunks, metadata & vectors. Vector search and optional hybrid search.
Native reranking
Managed reranking of candidate passages before evidence assembly.
Retrieved evidence returns to the Gateway before the prompt is sent to Ollama. Atlas serves the RAG knowledge base; the RK1 cluster supports separate database experiments.
03 / THE RK1 DATABASE LAB
Three database members. A dedicated control node.
The public MongoDB project separates the replica set from the tools used to deploy, observe, and test it.
MONGODB REPLICA SET
*Illustrative roles. A replica-set election can change which member is primary.
04 / DOMOTICS & HOME AUTOMATION
A connected home. One control layer.
Home Assistant brings together home monitoring and control across Z-Wave, LoRa, Zigbee, and Wi-Fi. An MCP connection links this home automation layer to Hermes.
Agent interaction with home automation tools
Model Context Protocol connection
Monitoring, automations & control
PROTOCOLS & INTEGRATIONS
Devices connect through the appropriate integrations and gateways.
Pool
Monitoring & control
Temperature
Measurements & trends
Air quality
Visibility into indoor conditions
Heating & cooling
Climate monitoring & control
Power monitoring
Understand energy consumption
Power optimization
Coordinate energy use to help control costs
Monitoring makes conditions and consumption visible. Automations coordinate home systems and optimize power use to help manage running costs.
Make it repeatable.
Turn setup steps into automation. Keep the configuration understandable and make rebuilding part of the learning process.
Measure the system.
Use observability and controlled experiments to distinguish a promising idea from a demonstrated result.
Keep asking questions.
Follow a build into the next experiment: from running MongoDB on ARM to understanding memory, and from a local model to grounded answers.
HOW THE IDEAS CONNECT
Infrastructure as
a learning tool.
The public projects explore three complementary layers. This map describes those themes, rather than a deployed network.
Sharing the learning.
This notebook publishes conceptual architecture, public project details, and selected results. Internal addresses, hostnames, access paths, private documents, and operational dashboards stay private.
Read the approach →