01 / LOCAL AIOllama at the edge
Evaluate local models for voice, vision, coding, automation, and private document context on practical home hardware.
Smart Home Edge Systems / Working lab
Research that becomes useful at home.
The R&D hub is where prototypes become dependable deployments. We test local AI, voice pipelines, private networking, and resilient infrastructure in an in-home lab before documenting what works.
01 / Active direction
From experiment to evidence.
R&D connects the physical testbed to the guides and benchmark matrix published across the site.
01 / LOCAL AI
02 / Multimedia on the EdgeMulti-Media Generation on the Edge
Use Local AI models and Local tools to create pictures, audio and integrated video without using web-based tools.
03 / VOICE
End-to-end voice testing
Trace microphone capture, wake word, STT, Home Assistant intent, Piper TTS, network transport, and playback.
04 / INFRASTRUCTURE
Private system design
Build repeatable Proxmox VM/LXC patterns with isolated backend networks, secure access, and dependable recovery paths.
05 / DNS
Unbound recursive DNS
Pair privacy-focused recursive resolution with Pi-hole or AdGuard Home for local filtering without routine dependence on public DNS providers.
06 / FILESSamba file sharing
Test fast, browsable cross-platform file access for media libraries, pictures, backups, and the applications that use them.
07 / MEDIAHome media services
Compare Plex, Jellyfin, and Emby for organizing and streaming local movie, television, and music collections.
08 / AUDIOAudiobookshelf
Evaluate dedicated self-hosted audiobook and podcast streaming with household-friendly access and library management.
09 / CLOUDPrivate cloud storage
Assess Nextcloud and ownCloud as local alternatives for file sync, calendars, documents, and collaboration.
10 / PHOTOSImmich photo archive
Test mobile auto-upload, local photo and video backup, search, and AI-assisted face detection within the private network.
11 / DOCUMENTSPaperless-ngx
Build a searchable household archive through scanning, OCR, indexing, retention, and document retrieval.
12 / SECURITYVaultwarden
Evaluate a lightweight, self-hosted Bitwarden-compatible backend for private household credential management.
13 / DATASystem database
Compare PostgreSQL, MySQL, and SQLite for service metadata, automation records, analytics, and operational reporting.
14 / RETRIEVALDeluge file retriever
Test controlled torrent retrieval with clear storage boundaries, service isolation, and responsible network policies.
15 / FABRICATIONManyfold / Printstash
Index, tag, and retrieve 3D-printing files so models, parts, revisions, and materials remain useful over time.
16 / WORKSPACEDIY test workspace
Maintain a hands-on area for voice, media servers, file sharing, Immich, Vaultwarden, and hardware integration trials. Also provide a mechanism to test the level of security and be able to predict and counterattack threats as they occur.
17 / OPERATIONSGitHub issue tracker
Use issues and project boards to record defects, experiments, sponsorship tasks, decisions, and release milestones.
18 / OPERATIONSArcade and Gaming Server
Provides a system with extensive GPU that will drive many of the latest owned Arcade and Gaming systems.
19 / Sentinal ToolsServer Monitoring & Application Monitoring (APM)
Deliver a modern monitoring and incident-management platform that fits both personal projects and growing teams. It bundles uptime checks, heartbeats, Slack and email alerts, exception tracking, log storage, and metric retention into a simple, fast interface.
20 / Virus ToolsData Scanning and Virus Control
Deliver a modern monitoring and incident-management platform that will keep track of the mass storage funtions.
21 / BUSINESSBusiness and environment tools
Connect analytics, calendar items, a personal weather station, an astronomical clock, and Xcel smart-meter measurements to Home Assistant and the lab record.
02 / Design implementation requirements
A testbed with standards.
These requirements keep the work affordable, honest, and useful to both households and potential technology partners.
Private by default
Prefer local processing, document network boundaries, and verify idle behavior rather than assuming privacy from product claims.
Repeatable builds
Record hardware, versions, configuration, resource allocation, and test conditions so results can be reproduced.
Measured performance
Track latency, accuracy, throughput, memory, CPU/GPU use, power, and failure behavior across the full workflow.
Accessible cost
Favor inexpensive, repairable, well-documented components and explain the tradeoffs behind each recommendation.
Open documentation
Publish implementation guides, test notes, references, and limitations so the community can inspect and improve the work.
Responsible partnerships
Accept sponsored hardware or software for evaluation while keeping test conditions and results transparent.
03 / Current AI workload
Models with a job to do.
Each model is selected for a defined role in the local system, not simply because it is new or popular.
gemma3:4b
Vision and general assistance.
phi4-mini
Reasoning and tool calling.
qwen2.5-coder:7b
Automation engineering and YAML.
qwen2.5-coder:1.5b
Inline LAN autocomplete.
nomic-embed-text
Local document retrieval.
04 / Working business model
Build trust before scale.
Partner with the testbed.
The R&D program is designed to grow through product sponsorship and practical collaboration. Hardware and software partners provide evaluation access; the lab provides real-world testing, documented configurations, and community-facing results.
Support can help fund:
- Hardware units and replacement components
- Software access and developer tooling
- Measurement, logging, and media production
- Open guides and benchmark publication
- Community demonstrations and field trials
05 / Continue the work