Landlord builds fully local AI pipeline to turn property inspection videos into Trello repair lists

gregmushen shared an automation project built to solve a real pain point: he owns a property, and each inspection means recording a 25-30 minute video, then manually going through it, taking screenshots, and logging repair items into Trello—very time-consuming. So he built a fully local AI pipeline that automatically extracts repair screenshots from walkthrough videos and generates Trello repair to-dos; each run takes about 2.5 hours, with no cloud models involved.

Confirmed

• Tech stack: Silero VAD v5.1.2 first strips silent segments to prevent Whisper from hallucinating on silence; Whisper-Large-v3-turbo then handles speech transcription; Gemma 3 8B (Q4KM quantization) extracts repair-related information
• Output: an organized list of repair items; clicking any item shows the repair details, corresponding screenshots, and a direct jump to the video segment discussing the issue
• Development approach: specs and implementation plan written with Astra High, reviewed by Opus 5.5, then built end-to-end with tests; Opus 5.5 acted as orchestrator, delegating development to multiple sub-agents

Why it matters

• Demonstrates a reusable video-understanding pipeline of "VAD preprocessing + speech transcription + local small-model structured extraction"; handling silence-induced hallucinations is a practical engineering detail
• Fully local execution means privacy-sensitive property/inspection footage never leaves the machine
• The multi-agent division of labor in development (planning, review, sub-agent execution) is itself a noteworthy AI-assisted development paradigm

2026-10-09 ~ 2026-10-09 · 11 related posts

• [source] Dev builds local-AI tool that turns property walkthrough videos into repair lists — gregmushen · 2026-10-09
• Landlord builds AI tool to auto-extract repair screenshots from walkthrough videos — gregmushen · 2026-10-09
• Multi-agent workflow: plan with Astra, review and orchestrate via Opus 5.5 subagents — gregmushen · 2026-10-09
• Forgot to enable remote mode? Have Codex use computer use to fix it — gregmushen · 2026-10-09
• [source] Homeowner builds fully local AI pipeline that turns walkthrough videos into Trello repair lists — gregmushen · 2026-10-09
• 2.5 hours fully local, plus Codex computer-use to fix a remote-control slip — gregmushen · 2026-10-09
• [source] Local pipeline: Silero VAD + Whisper + Gemma 8B in Ollama for video extraction — gregmushen · 2026-10-09
• Inside the repair review dashboard: default styling, very 'Claude'-ish — gregmushen · 2026-10-09
• AI taking over mundane life: it saves not just time but stress — gregmushen · 2026-10-09
• Repair dashboard links each fix to the exact video timestamp — gregmushen · 2026-10-09
• 'Insanely overkill' — yet the AI repair tool saves an immense amount of time — gregmushen · 2026-10-09