Inspiration
Smart home devices are notoriously insecure, often shipping with default credentials, unpatched vulnerabilities, or end-of-life software. We wanted to build an automated "watchdog" that doesn't just warn you about these risks, but proactively tracks your hardware and instantly locks it down if it starts acting suspicious.
What it does
Casawarden acts as an automated security layer for smart home networks. It continuously monitors your network to build a live inventory of your hardware and scores each device's risk based on known vulnerabilities and default passwords. If a device starts acting suspiciously (detected by using a sliding 10-second window to watch for unauthorized ARP sweeps or interactions with decoy targets) Casawarden instantly quarantines it. It then passes the threat data to the Gemini API, which translates complex anomaly alerts into user-friendly dashboard notifications with actionable next steps in plain English.
How we built it
Casawarden consists of a Python-based security agent designed to run on a Raspberry Pi, paired with a modern frontend dashboard. The backend evaluates devices based on static risks and dynamic network behavior. We built the frontend using React, Vite, and Tailwind CSS, polling the backend to display a live event feed, device risk bands, and isolation statuses. To make the alerts accessible, we integrated the Gemini API to generate structured JSON outputs containing plain-language explanations and cleanup steps for quarantined devices.
Challenges we ran into
One of our biggest hurdles was figuring out how to test and demonstrate network attacks without requiring the judges or our teammates to have physical, vulnerable IoT hardware on hand. We solved this by engineering a simulator that uses mock fixtures to post scripted attacks to our ingest endpoint, allowing anyone to test the dashboard's reaction to an attack in real-time.
Accomplishments that we're proud of
We are particularly proud of our fail-safe AI architecture. By enforcing strict schema validation with the Gemini API, we reliably parsed the responses directly into our React frontend without breaking the UI. If the API request ever fails or times out, the system immediately falls back to hardcoded template text, ensuring the core security of the application remains completely reliable.
What we learned
We learned a massive amount about network traffic analysis, specifically how to implement sliding windows for observing ARP requests. We also learned valuable lessons in system architecture, particularly how to build resilient integrations where generative AI enhances the user experience without acting as a single point of failure.
What's next for Casawarden
Moving forward, we want to expand our threat detection capabilities to cover more complex IoT protocols (like Zigbee and Z-Wave) and deepen our direct integration with Home Assistant. We also plan to enhance our simulator to include a wider variety of automated attack vectors for better stress-testing and demonstration purposes.
