Daniel Cárdenas

AI Automation Engineer: RAG · Edge ML · Firmware

I build practical AI systems: RAG pipelines, voice and document automation, Python APIs, and the embedded telemetry that makes edge ML useful in the real world.

XPLevel 1
0 XP150 to level up
Aug 2025•Backend / Computer Vision Developer•2 min read

Wyze Lab Monitor

FastAPI service that captures Wyze Bridge camera snapshots, uses Gemini OCR to read lab thermometer/hygrometer displays, and stores environmental history with alerts and correction tools.

Gemini OCRWyze BridgeSQLiteAlerts
PythonFastAPIGeminiComputer VisionDockerSQLiteIoT

Confidential Context

This case study is sanitized. Client data and proprietary integrations are omitted. Internal lab monitoring project; camera credentials, endpoints, and raw environment data are private, so this case study is anonymized.

Outcomes

  • Automated 15-minute captures from Lab and BodLab cameras
  • Merged camera-derived readings with RHT-20 datalogger uploads for Bodega
  • Stored 20k+ local readings with charting, image history, alerts, and manual correction flow
  • Added Docker Compose operations around Wyze Bridge, FastAPI, startup, reset, and health scripts

Problem

The lab needed regular temperature and humidity history for environmental monitoring, but the available instruments were a mix of camera-visible UNI-T A12T thermometer/hygrometers and an RHT-20 datalogger.

The practical problem was not just reading values once. The system needed to keep collecting data, survive camera/bridge instability, show history, preserve image evidence, and let a human correct OCR mistakes without losing the audit trail.

Approach

  • Camera ingestion: used Wyze Bridge as the local camera source and scheduled FastAPI captures every 15 minutes for Lab and BodLab.
  • Structured OCR: sent each snapshot to Gemini with a strict JSON schema for indoor temperature, outdoor temperature, humidity, and display time.
  • Local persistence: stored readings in SQLite with uniqueness by sensor/timestamp and retained recent camera images for review.
  • Mixed sensor inputs: accepted RHT-20 datalogger uploads for the Bodega sensor through a dedicated API endpoint.
  • Dashboard: built a browser UI with live values, Chart.js history by sensor/date/metric, alert threshold configuration, and image review.
  • Correction workflow: chart points can open the captured image and submit corrected temperature/humidity values back to the database.
  • Operations: added Docker Compose for Wyze Bridge + FastAPI, startup scripts, bridge reset scripts, health checks, and a nightly recovery path.

Results

  • Continuous 15-minute environmental capture for camera-backed lab sensors.
  • 20k+ local readings across Lab, BodLab, and Bodega during the reviewed data window.
  • Fast inspection of historical readings with the original image attached to camera-derived samples.
  • Lower operational friction through reset/startup scripts for the Wyze Bridge and sensor app.

Skills Demonstrated

  • Practical computer vision/OCR workflow using camera snapshots and structured LLM output.
  • FastAPI API design for capture, history, alerts, corrections, image retrieval, and datalogger uploads.
  • Docker-based local operations for an IoT monitoring stack.
  • Product sense around human-in-the-loop correction where OCR can be wrong.