Personal Project
PRODUCT MANAGEMENT
One night, I got up early to go to the gym. As I walked past the front of the house, something felt off. Our garage door was wide open. I walked further and noticed the side gate into our backyard was also open. When I went into the garage, it became clear what had happened. My eScooter was gone. We had been broken into.
What unsettled me most was not just the theft, but the realisation that we had not noticed anything while it was happening. I am generally a light sleeper, but that night it was windy. We did have security cameras, but reviewing footage after the fact did nothing to deter the intruders or help us feel safe in our own home.
In the days that followed, I felt stressed, exposed, and uneasy. We love our home and our neighbourhood, so moving was never an option. The problem became clear: how do you turn a house from something that simply records incidents into something that actively protects the people living inside it?
Problem framing through lived experience
Systems thinking and solution design
Hardware and infrastructure research
AI and computer vision experimentation
Automation architecture
Network configuration and troubleshooting
Starting small and falling down the rabbit hole
Before the break-in, my wife and I had only dabbled in basic smart home features. Automated lights. Voice commands. Speakers that played music in different rooms. Useful, but shallow.
I had heard of Home Assistant before and knew it was powerful, but it always felt like something for people far more technical than me. At the same time, my day job involved working on a complex automation project, and I genuinely enjoyed that work. The break-in became the catalyst to finally explore what was possible.
We started modestly with a Raspberry Pi and an AI acceleration HAT. The initial goal was simple: make our security cameras actually do something.
Turning cameras into active sensors
We discovered Frigate, a tool that allows camera feeds to be processed by an AI model capable of detecting human presence. This was a critical breakthrough. Motion sensors were not good enough. We did not want to be woken up at 2am because a possum walked past or a bug flew close to the lens.
By feeding our camera streams into Frigate and integrating it with Home Assistant, a “person detected” event became a reliable trigger for automations. This completely changed how we thought about security.
We designed our first deterrent automation so that if a person was detected during certain hours of the night, the lights inside the house would turn on in a staggered sequence, making it look like someone had woken up and was moving through the house. I also built a Discord bot that notified us immediately so I could get up and monitor the situation in real time.
A few months later, this setup proved its value. Two young kids were detected snooping around cars in our driveway. The lights came on. They ran away.
It worked. But it was also confronting. We realised we wanted something stronger.
Scaling the system and designing for escalation
At this point, the limitations of our hardware became clear. The Raspberry Pi struggled to run AI detection on more than a single camera. If this system was going to be reliable, it needed more compute.
I spent a significant amount of time researching hardware options, learning about CPU cores, memory constraints, and inference workloads. Eventually, we upgraded to a Mini PC capable of running AI detection across all five of our security cameras simultaneously.
With the extra headroom, we redesigned the system around layered responses:
Level 1: If a person is detected in the driveway, lights stagger on inside the house.
Level 2: If a person is detected near the front door or side of the house, a siren activates and speakers announce “Intruder Alert”.
This escalation model balanced deterrence with restraint and reduced the risk of unnecessary disruption.
Going deeper than we ever expected
As the system grew, so did its complexity.
Networking became another challenge entirely. Camera streams, local IPs, bandwidth constraints, and reliability issues were all far outside my comfort zone as a product manager. Progress often meant hours of trial, error, and debugging. But each breakthrough reinforced the value of designing systems properly rather than relying on quick fixes.
Hardware choices became increasingly deliberate. We prioritised IoT devices that were reliable, scalable, and capable of running locally. Zigbee was chosen as a core protocol so the system could continue operating even if the internet went down, removing cloud dependencies entirely.
From security system to intelligent home
While the obsession started with security, something unexpected happened. Once we had a stable platform, we began seeing opportunities everywhere.
We automated our laundry fan so it only runs when the dryer is on. We motorised our roller blinds so they open automatically when we stop our morning alarm. We built a sophisticated lawn watering system that tracks soil moisture and weather forecasts to water only when truly needed. We created a “Netflix & chill” mode where lights dim and blinds lower automatically in the evening.
We also leaned into smart notifications. Messages when the washing is finished. Alerts when a battery is running low. Warnings when an automation fails or when the garage door has been left open.
Today, our home runs over 50+ automations.
What started as a response to a break-in became one of the most rewarding personal projects I have worked on.
We feel significantly safer in our home, knowing it actively responds to threats rather than passively recording them. At the same time, our day-to-day life is easier, calmer, and occasionally delightful. I still cannot get over the feeling of the blinds opening as we wake up, or the lights softly coming back on after a movie ends.
Beyond the home itself, this project demonstrates how I approach complex problems: start with a real human need, learn relentlessly, design systems thoughtfully, and iterate until the solution feels both robust and humane.