Week 16 - System Integration

Individual assignment

This week was about turning separate tests into one object. My final project, Brain Fog Insight Companion, has a desktop host and a wearable band. Before this week I had working pieces on breadboards: screen, speaker, microphone, camera, local eye-state model, BLE band, cloud voice, and 40 Hz audio. The system integration job was to decide where each part lives, how power and data move, and how to make the prototype look like one finished device instead of a pile of modules.

The project is a wellness and behavior-support prototype. It is not a medical diagnostic device.

Integration plan

I planned the product as a laser-cut platform with electronics inside. The left side holds the 3D printed host shell, with the screen and speaker fitted into the front. The right side holds a small laser-cut tower for the camera. The front edge reserves a magnetic charging position for the wearable band. A left push button triggers proactive voice dialogue. A right master switch controls the desktop host, while the watch charging line stays available. A Type-C port on the back powers the platform.

Hand sketch of Brain Fog Insight Companion system layout — Planning evidence: sketch, 2D laser-cut platform drawings, cutting process, and 3D host-shell model used before physical integration.

System architecture

The final architecture uses more than one microcontroller because each board has a narrow job. That made debugging less confusing. The WROOM host does the screen, audio, voice, and main state logic. The XIAO ESP32S3 Sense runs the local eye-open / eye-closed model from its camera and reports blink state to the WROOM through GPIO. The wearable band sends heart-rate and SpO2 data over BLE. A XIAO ESP32-C3 bridge listens for those advertisements and forwards clean lines to the WROOM over UART.

Subsystem	Connection	Why it is separate
Wearable band	BLE advertisement to XIAO ESP32-C3 bridge	It can stay on the wrist and send HR / SpO2 without pairing delays.
XIAO ESP32-C3 bridge	UART to WROOM host	The WROOM does not spend its main loop managing BLE scan details.
XIAO ESP32S3 Sense	Camera model output through GPIO to WROOM	Eye open / closed detection stays local and simple: the host receives state, not video.
WROOM host	Screen, speaker, microphone, SD card, WiFi, buttons, power switch	It owns the user-facing loop: display, voice, intervention, and logging.
Alibaba Bailian voice path	WiFi WebSocket from WROOM	Used for ASR, LLM reply, and TTS during voice interaction.

End-to-end function flow

The finished prototype is meant to run as one desk device, not as separate electronics tests. A normal session starts from the rear Type-C power input. The right switch powers the desktop host, while the wearable charging position on the front can still be used. The screen boots into the host UI, the camera tower watches the user's eye state through the XIAO ESP32S3 Sense, and the wearable band sends heart-rate and SpO2 data through the C3 bridge.

During use, the WROOM combines three inputs: blink state from the XIAO S3, biometric data from the wearable, and optional voice input from the microphone. The screen shows the current state and the system can run an intervention: 40 Hz audio, a guided breathing or behavior suggestion, or a spoken response from the cloud voice path. The left button gives the user a direct way to start proactive dialogue without touching code or opening a serial monitor.

Step	What happens	Hardware involved
1. Power and wake	Rear Type-C powers the platform; the right switch turns on the desktop host.	Power input, master switch, WROOM host
2. Sense	Camera classifies eye open / closed; wearable measures HR and SpO2.	XIAO ESP32S3 Sense, Beetle band, MAX30102
3. Transfer	Eye state reaches WROOM through GPIO; wearable data reaches WROOM through C3 UART.	XIAO S3 GPIO link, XIAO C3 BLE bridge
4. Decide	WROOM updates the UI and decides whether the state looks normal or needs intervention.	WROOM, screen, firmware logic
5. Intervene	The system can play 40 Hz audio, show guidance, or answer through voice interaction.	Screen, microphone, MAX98357A speaker path, WiFi voice service
6. Re-check	The same sensors keep running so the user can compare the state before and after.	Camera, wearable, WROOM UI

Firmware flow

The product runs four small programs at once. Each board keeps one job so the host is not doing everything in one loop.

Board	What the program does
Beetle band	Reads the MAX30102 about once per second, then sends heart rate, SpO₂, and temperature over BLE.
XIAO ESP32-C3 bridge	Scans for the band, parses each packet, and forwards JSON lines plus a link-status message to the WROOM over UART.
XIAO ESP32S3 Sense	Runs the local eye open/closed model and reports eye state to the host. The WROOM never processes camera frames.
WROOM host	Owns the touch UI, 60-second measurement session, risk score, intervention playback, SD history, and cloud voice.

On the host, a normal user session looks like this:

Boot and home. After power-on the screen shows live vitals, band link status, and current eye state. The host keeps reading UART lines from the C3 bridge and eye-state updates from the camera board.
Start detection. The user starts a 60-second measurement from the touch UI. During that minute the host logs one sample per second: blink rate from the eye model, plus the latest heart rate and SpO₂ from the band.
Count blinks. The camera board only reports open, closed, or transition. The host runs a small debounce step so a single slow close does not become ten blinks. Only a full open → close → open cycle counts.
Result page. When the minute ends, the host combines blink rate, heart rate, and SpO₂ into one rule-based Brain Fog score from 0 to 100, then maps it to Low, Mild, Moderate, or High. A short text suggestion appears on screen.
Intervention offer. The host asks whether to run an intervention. If the user accepts, it plays about three minutes of 40 Hz audio through the speaker and flashes a green area on the screen. The user can stop early. When playback ends, a short post-intervention message plays.
Voice on demand. The left button starts a cloud voice session at any time: microphone capture, speech recognition, Aquarius reply, then text-to-speech on the speaker. Voice and measurement can block each other so only one runs at a time.
History and standby. Finished sessions can be stored on the SD card and reviewed later. If nothing happens for a while, the screen can dim to a sleep page until the user taps to wake it.

The score is rule-based, not a trained medical model. Blink rate carries the most weight because it is always available at the desk. Heart rate and SpO₂ adjust the score when the band is connected and the sensor has good skin contact.

WROOM wiring map

I kept a pin map while moving from breadboard to PCB. This table is the wiring used for the integrated desktop host. The C3 bridge wiring also matches the bridge firmware comments: C3 D6 is TX to WROOM GPIO39, and C3 D7 is RX from WROOM GPIO38.

Module	Module signal	WROOM / power connection
Screen + SPI adapter	NT / INT	GPIO14
	NC	Not connected
	SDA	GPIO13
	RST	GPIO12
	SCL	GPIO11
	SDO / MISO	GPIO10
	LED	GPIO9
	SCK	GPIO46
	SDI / MOSI	GPIO3
	DC	GPIO8
	RESET	GPIO18
	CS	GPIO17
	GND	GND
	VCC	3V3
INMP441 microphone	SCK	GPIO6
	WS	GPIO5
	L/R	GND
	SD	GPIO1
	VDD	3V3
	GND	GND
XIAO ESP32S3 Sense	3V3	3V3
	GND	GND
	D6	GPIO16
	D7	GPIO15
Proactive dialogue button	Signal	GPIO7
MicroSD module	VCC	3V3
	GND	GND
	CS	GPIO2
	MOSI	GPIO42
	SCK	GPIO41
	MISO	GPIO40
	INST	Not connected
XIAO ESP32-C3 bridge	D6 / TX	GPIO39 / RX
	D7 / RX	GPIO38 / TX
	GND	GND
	3V3	3V3
MAX98357A playback module	LRC	GPIO48
	BCLK	GPIO47
	DIN	GPIO21
	GAIN	Not connected
	SD	Not connected in firmware pin map
	GND	GND
	VIN	5V

Wearable wiring

The wearable band uses a Beetle ESP32-C6 and MAX30102 heart-rate / SpO2 module. The Beetle firmware starts I2C on GPIO6 and GPIO7, reads the sensor at address 0x57, then advertises seq / hr / spo2 / temp as BLE manufacturer data.

Part	Signal	Connection
MAX30102 sensor	SDA	Beetle GPIO6
	SCL	Beetle GPIO7
	VCC	3V3
	GND	GND
Beetle ESP32-C6	Battery	LiPo through the Beetle charging circuit
Beetle ESP32-C6	Status LED test	GPIO15 in the test firmware

Packaging

The packaging moved in two stages. First I proved the circuit on a breadboard. Then I moved the host electronics onto my own PCB so the desktop unit could be placed inside the laser-cut platform. The PCB step mattered because the breadboard version worked, but it was too fragile to close into a product-like base.

Breadboard final hardware test — Final breadboard test. This was the last loose setup before moving the host electronics to the custom PCB.

KiCad layout of final host PCB — KiCad layout for the host PCB. I added this revision after the breadboard wiring was stable enough to commit.

Assembled host PCB — Host electronics soldered onto my PCB. This replaced the breadboard for the main unit.

Internal circuit routing inside the laser-cut platform: screen ribbon, buttons, charging contacts, and power wiring before closing.

Inside the platform before bottom cover — Inside the platform before closing the bottom. This is where I checked wiring, button placement, and power routing.

Physical layout

The desktop product is built like a small base station. The laser-cut base carries the load and hides the wiring. The 3D printed shell on the left gives the screen and speaker a fixed home. The camera tower on the right holds the camera at a better viewing height than the base alone. The front charging position keeps the wearable with the host, so the two-device system does not feel like two unrelated prototypes.

3D model of host shell front — 3D shell planning. I had already learned from earlier prints that the screen board needs more space than the front glass size suggests.

3D printed host shell opened — Host shell before final fitting. The screen and cable path drove the internal clearance.

Wearable band prototype with battery and sensor. The desktop base has a reserved charging position for it.

Final integrated desktop product — Integrated product form: host shell on the left, camera tower on the right, controls and charging area on the front.

Integration test

I tested the whole platform before the bottom cover was closed. This video was not meant to be pretty. It was a practical check: power up, screen on, internal wiring still visible, and the main loop running before I hid the electronics. Catching failures at this stage is easier than reopening a finished platform.

System test before closing the base. I used this run to check whether the integrated desktop unit could boot and run with the wiring in its final positions.

What changed during integration

Issue	Change	Result
Breadboard wiring was too loose for a closed platform.	Moved the host electronics to my custom PCB.	The wiring became shorter and easier to mount.
The camera needed a stable view of the user.	Added a laser-cut tower on the right side of the platform.	The camera is no longer buried at table height.
The wearable could become a separate object and get lost.	Reserved a magnetic charging position at the front.	The wearable belongs to the same physical system as the host.
The user needs a quick way to start voice interaction.	Added a front button for proactive dialogue.	The main interaction is available without opening a debug terminal.
The host needs a hard power control, but charging should remain possible.	Separated the master switch behavior from the watch charging line.	The desktop host can be switched off while the wearable still charges.

Repair and access

The platform is not sealed permanently. I kept access to the bottom because the system still needs service: firmware uploads, cable changes, and sensor replacement are normal for this prototype. The Type-C input is on the back so power does not compete with the front controls. The next mechanical improvement would be better strain relief for the internal cables and a cleaner removable bottom panel.

Design source files

The host-shell 3D design is the same enclosure family documented in Week 2 CAD and Week 5 3D printing. For this integration week I added the final electronics source package for the Brain Fog host PCB:

Download Brainfog KiCad source package - project file, schematic, PCB layout, custom footprints, and DRC/ERC reports.
3D host shell design process - same enclosure source family used for the final host shell.

Link to final project

This page documents the system-integration step for the Brain Fog Insight Companion final project. The final project page links back to this week as the packaging and integration record.