Ghost in the Machine: my smart home writes down what happens after dark

Description and use

The system turns smart‑home sensor logs into a nighttime story, detecting anomalies locally and narrating them aloud.

Problem

How to turn a smart home’s sensor logs into a narrative that reveals unusual nighttime events without exposing private data.

Outcome

The system accurately flags unusual events (e.g., bedroom door opening at 02:19 while both occupants are still in bed) and produces a coherent, timestamped story that can be listened to or viewed in a replay UI.

Workflow

• Export Home Assistant history to a CSV file
• Run TabPFN to score each event for anomaly using a 40‑night simulated baseline
• Flag events with score ≥0.95 as potential ghosts
• Feed flagged events to Gemma 4 with a prompt that requires citing real log lines
• Gemma 4 outputs a narrative with timestamps and sensor names
• Piper TTS reads the narrative aloud
• Optional: replay the log in the web UI highlighting anomalous lines

Evidence

TabPFN finds the ghosts. A real Home Assistant export doesn’t come labelled. Each event becomes a row of generic features… TabPFN, an open‑weight tabular foundation model, takes 40 labelled simulated nights as context and scores a new night in a single forward pass, with no training loop.

Source

Gemma 4 tells the story. The flagged readings go to the narrator tagged [anomaly 0.97]. The prompt’s key rule: every eerie beat must cite a real log line by time and sensor, and the model may not invent readings.

Source

Piper reads it aloud. Open‑source neural TTS, fully offline, slowed down a little and with times spoken as words.

Source