Homelab
24 self-hosted services, 80+ automated workflows, and a local AI stack. All running on-premise. No cloud dependencies, no third-party data processing.
Architecture
Hardware
TerraMaster F2-424
16 GB RAM, ZFS
20+ Docker containers
Windows PC
Ryzen 5 5600X, 32 GB RAM, RTX 4060
LM Studio, WhisperX, Ollama
Home Assistant Green
Dedicated HA appliance
Home Assistant + voice satellites
MacBook Pro
Development, Obsidian sync
Obsidian REST API
Services (24)
Workflow automation: 80+ workflows handling integrations, pipelines, and scheduled tasks
Smart home control with voice satellites and device automation
Local LLM inference (Qwen3.5 9B) with OpenAI-compatible API
GPU-accelerated speech-to-text with speaker diarization for transcription pipelines
Secondary LLM backend for model evaluation and testing
RAG interface for querying documents with conversational search (384-dim embeddings)
Vector database powering semantic search across all stored content (384 dimensions)
Document ingestion with OCR, tagging, and full-text search
Internal wiki for journals, runbooks, and reference documentation
Primary note-taking and knowledge management via Markdown
RSS aggregation with AI-scored relevance filtering
Bookmark manager with full-text archival and tagging
Media server for movies, TV, and music libraries
Self-hosted photo management with facial recognition and search
Bitwarden-compatible password manager, fully self-hosted
VPN for secure remote access to all internal services
Cloudflare tunnels for externally exposed services
Service health monitoring with alerting via ntfy
Real-time system metrics: CPU, RAM, disk, and network
Push notification hub for all automated alerts
Local PDF processing: merge, split, convert, OCR
File sync across reMarkable, Obsidian vault, and workstations
Self-hosted remote desktop access
Privacy-first budgeting and financial tracking
Automation Highlights
80+ n8n workflows handle recurring tasks, data pipelines, and infrastructure maintenance.
Morning Briefing
Daily summary delivered by email: weather, calendar, tasks, and Oura ring sleep data, generated by the local LLM.
Transcription Pipeline
Omi wearable captures conversations, WhisperX transcribes locally with speaker diarization, n8n classifies content and routes action items to Todoist.
Personal Knowledge Management Pipeline
Notes from Obsidian, scans from reMarkable, and documents from Paperless are processed, embedded, and indexed for semantic search.
Self-Healing Infrastructure
Downtime triggers AI triage, automatic restart attempts, and rollback if health checks fail. Alerts sent via ntfy.
Docker Update Pipeline
Weekly automated container updates with pre-update snapshots, health checks, and automatic rollback on failure.
RSS AI Digest
FreshRSS feeds scored by relevance, top articles summarized by the LLM, and a daily digest pushed to ntfy.
AI Stack
Local LLM Inference
LM Studio serves Qwen3.5 9B on an RTX 4060 via an OpenAI-compatible API. All AI workflows (classification, summarization, briefings) route through this single endpoint.
RAG & Semantic Search
Documents, notes, and scans are embedded into Qdrant (384-dim vectors). AnythingLLM provides a conversational interface for querying the full knowledge base.
Speech-to-Text
WhisperX on the GPU handles transcription with speaker diarization for the Omi wearable and Home Assistant voice satellites. All processing stays local.