I wanted to share a medical IoT proof-of-concept I recently built for a hackathon using the ESP32. It’s called BEAM (Bilirubin Evaluation and Analysis Meter).Clinical-grade Transcutaneous Bilirubin (TcB) meters cost thousands of dollars, making non-invasive jaundice screening inaccessible in low-resource settings. BEAM is an attempt to replicate the core optical physics of those commercial devices at a fraction of the cost, while leveraging the ESP32 to push real-time data to a Next.js web dashboard.
Note: This is an open-source experimental prototype for educational/demonstration purposes only, not an FDA-approved diagnostic tool!
The Hardware StackMCU: ESP32 Development Board (NodeMCU-32S)
Sensor: SparkFun AS7343 14-Channel Spectral Sensor (I2C)
Display: 1.8" ST7735 SPI TFT LCD
Excitation Source: 5mm Cold White LED (GPIO 26)
Input: Tactile Push Button with software debounce (GPIO 27)
The Physics & MethodologyBilirubin is a yellow-orange pigment that heavily absorbs blue light (~450nm) but reflects red light. Instead of using a standard blue LED, the project uses a Cold White LED (which is fundamentally a blue diode coated in yellow phosphor). This gives us a massive emission spike in the blue range to excite the bilirubin, while providing a broad red/green baseline.The ESP32 reads the raw values from the AS7343's Blue (F3/F4) and Red (F7/F8) channels.
We then calculate the Optical Density (Absorbance) using a modified Beer-Lambert approximation for a 16-bit sensor:
$$A = \log_{10}\left(\frac{65535}{RawLight + 1}\right)$$
To account for varying skin melanin (which absorbs across the spectrum), we use a dual-wavelength subtraction model. The Red channel acts as our baseline, and the Blue channel acts as our target:
$$TcB = (A_{blue} \times M_1) - (A_{red} \times M_2)$$
As bilirubin concentration increases, blue light reflection drops, $A_{blue}$ spikes, and the final TcB score rises.
Firmware & Architecture HighlightsThe ESP32 firmware (written via Arduino Core) is juggling a few things to make this a seamless standalone device:
Display & UI: Uses Adafruit_GFX and Adafruit_ST7735 over SPI to render a custom boot screen, real-time raw data, a scan history array, and a visual "heartbeat" to ensure the main loop isn't blocking.Wireless Flashing: Integrated ArduinoOTA so I can update the math models and UI while the unit is completely sealed in its 3D-printed enclosure.
Cloud Sync: Upon a successful scan, the ESP32 packages the $A_{blue}$, $A_{red}$, and $TcB$ values into a JSON payload and fires a POST request via HTTPClient to a locally hosted Next.js API route.State Machine: Implemented strict hardware polling and software debouncing (50ms delay) so the sensor read cycle, LED firing, and HTTP requests only execute on an intentional state change.
Next Steps (ESP32-C3)For the next iteration, I plan to transition off the DevKit and design a custom PCB around an ESP32-C3 mini module to drastically miniaturize the form factor, moving the excitation LED to an SMD component placed as close to the sensor lens as possible.You can check out the full code, the calibration protocol (using standard yellow paper pigments to mimic bilirubin for judges!), and the Next.js dashboard here: 
https://github.com/KrishanuRoyEng/BEAM