Andrew Higgins

Anchored

An iPhone screen time app that hard-blocks different apps depending on which room of the house you are in, using a Bluetooth beacon in each room. Nobody approves their own extra time, parents included: they approve each other, and the whole family can see everyone’s usage. My solo entry for BYU’s Student Innovator of the Year 2027, now in the prototype stage.

• Designed room detection around BLE fingerprinting: a setup walk records each room’s signal signature, and live matching uses hysteresis and accelerometer gating so doorways don’t flicker.
• The policy is default deny: no signal, Bluetooth off or a dead beacon falls back to the strictest rules, and every unlock is a lease that expires unless fresh beacon sightings renew it.
• Approvals are planned as signed, expiring tokens the phone verifies offline, enforced through Apple’s FamilyControls and ManagedSettings frameworks.
• ESP32 room anchors with directional patch antennas, to be benchmarked against UWB and a single-beacon baseline for correct-room and doorway error rates.

Project Atlas

A four-layer STM32 control board built from scratch to replace the electronics controlling BYU's Formula SAE electric race car. The full manufacturing package is released and the first revision is fabricated and in hand. That board does not work yet, so the project is still open: rev 1 is where the 12 V rail died, and the fix belongs to rev 2.

• Designed the power tree: a fused 12 V input, a synchronous buck down to 5 V, and an LDO to 3.3 V.
• Six switched outputs off the STM32's GPIO: two 12 V high-side channels using PMOS switches with NMOS gate drivers, two open-drain 5 V channels, and two logic-level 3.3 V channels, with ADC sense on the low-voltage rails.
• Three independent FDCAN buses over TJA1051 transceivers, plus USB-C debug serial through a CP2102N bridge with ESD protection.
• Brought the first board up and traced a dead 12 V rail to a ground test point whose barrel crossed the split power plane.
• Released the full manufacturing package (four-layer gerbers, BOM, and pick-and-place) and set up the Git workflow a four-person team used to collaborate in KiCad.

Outlier Node

A battery-powered vibration sensor for machinery that has no network. It runs a machine-learning model on the device itself and sends only the verdict, not raw data, aiming at under $100 against commercial monitors that start around $275. Built with one teammate and shown at BYU's Student Innovator of the Year showcase.

• Designed the full circuit and PCB in KiCad (ESP32 Nano, MPU-6050 accelerometer, DS3231 real-time clock, and SD storage), and ran the bench bring-up and testing on every revision.
• Took the hardware from a breadboard proof-of-concept through four PCB revisions, using a coulomb counter to hunt down the power leakage that showed up on the early boards.
• Wrote the sensor firmware in embedded C++, including the low-power sentinel logic that keeps the sensor asleep until it needs to sample.
• Built the BLE transfer layer between sensor and phone, the “data mule” that offloads stored results as a technician walks past, so the sensor needs no Wi-Fi, cellular, or gateway.
• Specified AA lithium-thionyl-chloride cells for the multi-month battery target; the finished unit magnet-mounts to the machine in a sealed printed enclosure.

Wireless ECG

A four-node wireless ECG: sensor bands worn on the limbs plus a hub worn by the patient, replacing the wired leads of a clinical ECG while holding the four-lead configuration and 500 Hz sampling rate that FDA approval requires. A hackathon concept study, co-authored with two teammates. The pitch deck is complete and nothing is built.

• Did the system design and architecture, where time alignment was the hard part: an untethered node clock drifts about 50 µs every second, 13 ms across a five-minute test, which is fatal when four nodes have to agree on when a sample was taken.
• Had the hub rebroadcast its own clock fifty times a second and timestamp each sample in hardware rather than in software, dodging interrupt jitter and holding drift under one microsecond per sample. That supports sampling above 1000 Hz and roughly 48 hours of continuous operation.
• Solved the missing ground: a wired ECG gets its baseline from the right-leg electrode, which feeds an inverted signal back into the body to cancel ambient noise, and untethered nodes have no shared reference at all.
• Routed that reference through human body communication, using the skin itself as the conductive path so the body closes the analog circuit and ambient electromagnetic noise cancels in real time.
• Encrypted patient data under a key the nodes and hub exchange over a physical connection while charging, so the key never travels over the air and only the hub can decrypt.

Networked Smart Doorbell

A camera doorbell on an embedded Linux target, developed entirely over SSH against Makefiles I maintained by hand. Pressing the button captures a photograph, writes it to local storage, and uploads it to a web server, with a hidden file browser reachable through a button sequence.

• Worked around a display library that is not thread-safe: the capture and upload run on a POSIX thread while all drawing stays on the main thread.
• Moved images off the device over TCP sockets, and wrote a systemd unit that brings the program up on boot with a delay that keeps it from fighting the display for initialization.
• Built solo for BYU's ECEn 224 in C, on a Raspberry Pi Zero 2W, with no local editor or IDE on the target.
• The same course covered x86-64 assembly: reverse engineering a stripped binary in GDB and objdump, and stack-based exploitation including return-oriented programming.

Music Mood AI

Generates custom album art by analyzing the mood of a song and prompting an AI image generator.

• Analyzes audio features like tempo and timbre using Librosa and NumPy.
• Generates a descriptive "mood" prompt from the analysis.
• Uses OpenAI's DALL-E API to create a unique album cover.
• Displays results in a custom GUI built with Tkinter.

Home Infrastructure

An old laptop rebuilt as a headless Ubuntu server. It runs Jellyfin and Immich in Docker for media and photos, an OctoPrint instance for my 3D printer, and a self-hosted OpenClaw agent. Tailscale handles my own remote access, UFW rules sit at the edge, scheduled jobs cover backups and library scans, and storage is mounted and managed by hand.

Nothing here is novel. What it is good for is the part that does not show up in coursework: when a service stops at two in the morning, I am the one reading the logs.