Problem

A biotech lab relied on Wyze cameras pointed at instrument panels to prove temperature and reagent compliance. Interns manually transcribed readings, leading to gaps and late-night audits.

Approach

• Reverse-engineered Wyze camera RTSP feeds and deployed Raspberry Pi edge nodes to capture frames.
• Wrote OpenCV pipelines with Tesseract OCR tuned for seven-segment displays and printed labels.
• Aggregated readings via FastAPI, persisting to TimescaleDB and emitting alerts through Slack + SMS when drift exceeded thresholds.
• Added a calibration UI running as a PWA so technicians could re-align bounding boxes in seconds.
• Hardened stack with device watchdogs, systemd services, Prometheus exporters for uptime.

Results

• Fully automated capture every 5 minutes with SLA-backed alerts.
• Eliminated weekly compliance misses and improved audit readiness with downloadable CSV exports.
• Reduced manual transcription time by 6+ hours per week across shifts.

Links

• Source repository