AI for Ireland Quest 2026
I am thrilled to share that my project, FloodGuard AI, was awarded 2nd place at the AI for Ireland Quest 2026.
The hackathon was an incredible showcase of innovation across the country. To put the scale of the event into perspective, it featured:
600+ participants
40 final project submissions
900+ votes cast across the evaluation phases.
Standing on the podium alongside such top-tier talent is a massive honor. Here is a technical deep dive into how I built “FloodGuard AI - an autonomous flood defense system for Cork City” and the engineering decisions that secured the win.
The Vision: Moving Past Passive Dashboards
Current government flood monitoring tools are completely passive. They serve up water levels on a screen and rely entirely on a human operator to spot the danger, make manual phone calls, and coordinate a response. If a flash flood hits at 3:00 AM, those critical minutes wasted cost millions in damages.
FloodGuard AI changes the paradigm from “Human-in-the-Loop” to “Human-on-the-Loop.”
It is an autonomous agentic system built on a simple yet powerful triad: Sense, Think, and Act.
Sense: Continuously monitors real-time hydrometric networks.
Think: Uses Machine Learning to predict downstream flood risks 24 hours in advance.
Act: Autonomously executes defensive plays-like adjusting upstream dam discharges, rerouting traffic SCATS systems, alerting emergency services, and instantly broadcasting SMS warning logs to vulnerable local businesses.
The “40% Technical Leap” That Won the Podium
Midway through the event, the judges challenged participants to demonstrate a massive technical iteration from a basic mockup to a deployment-ready infrastructure candidate. That is where the real engineering began.
1. Training the Brain on 5 Years of History
I realized early on that real-time APIs only hold a rolling 35 days of data-nowhere near enough to train an intelligent predictive model. To solve this, I manually extracted 1,825 days of verified historical daily data (2021-2026) from the OPW Hydro-Data Archive for the Mercy Hospital station.
Using this dataset, I built an autoregressive Random Forest Regressor that looks at 3-day history lags, rate of rise, and weekly averages to accurately forecast tomorrow’s peak water levels.
2. Live API Infrastructure & The “Weir Narrative”
For live monitoring, my deployment build plugs directly into the live OPW network. I targeted the Waterworks Weir sensor (ID 19102) on Lee Road to read live river flow heading straight for the city center.
During my presentation, a sharp judge noticed my live app showed a water level of ~0.97m, while the public website reported ~2.96m. This discrepancy actually became my winning technical defense:
The website displays static headwater stored behind the weir wall. FloodGuard AI intentionally targets the downstream tailwater flow. Static water doesn’t flood homes-moving water does. The system monitors the actual threat, not the storage capacity.
3. Bulletproof Demo Planning: The Surge Slider
Because it was a safe, clear day in Cork during the evaluations, the live sensor feed sat at a completely normal, green level. To showcase the agent’s autonomous power, I implemented a Digital Twin Simulation Layer.
By adding a “Simulate Surge” slider on top of the live data stream, I could artificially introduce a 2-meter surge in front of the judges. Dragging that slider instantly turned the map crimson, generated automated HSE emergency evacuation registers, and activated secure CSV audit logs tracking the AI’s safety-critical decisions in real-time.
Lessons Learned & Next Steps
This competition reinforced that a winning AI project isn’t just about an attractive UI; it’s about data integrity, handling edge cases, and confidently explaining your architecture.
Because FloodGuard focuses heavily on critical civic infrastructure, I have implemented strict “All Rights Reserved / Evaluation Only” licensing in the repository to protect the intellectual property as I explore expanding this framework into a viable public-safety startup.
Check Out the Code
The entire implementation-including the live Streamlit dashboard layout, the 5-year training script, the live OPW data scraper, and the clean architecture folder structure is completely open for review.
🔗 Explore the project on GitHub: https://github.com/KourosZanbouri/FloodGuardAI
A massive thank you to the organizers of the AI for Ireland Quest 2026, the judges for grilling my architecture, and everyone who supported the project along the way! Next stop: turning this prototype into real-world protection.