Winter Is Coding Hackathon — Lessons from a Month of Large-Scale Optimization
From December 15, 2025 to January 15, 2026, nearly 850 students from EPITECH and other schools competed in the qualifying rounds of the Winter Is Coding hackathon.
A month of intense competition centered around a subject that was both technical, demanding, and deeply rooted in real industrial challenges.
The goal was clear: design a solution capable of optimizing 5G antenna deployment across a city, covering all buildings while minimizing total installation costs, despite strong constraints on range, capacity, and variable populations throughout the day.
A problem that, behind its apparent simplicity, actually hides a large-scale combinatorial optimization challenge, similar to the Facility Location Problem found in telecommunications, logistics, and industrial planning sectors.
A Problem Far More Complex Than It Appears
Each building had different population levels across three periods: peak hours, off-peak hours, and night. Antennas, meanwhile, had distinct ranges, capacities, and costs, with additional positioning constraints.
This wasn't simply about "placing antennas," but rather solving a problem combining:
- Optimal infrastructure placement (similar to the Set Cover Problem)
- Resource allocation under constraints
- Peak load management
- Constant trade-offs between cost, coverage, and capacity
Very quickly, one thing became evident: a naive or purely heuristic approach hit its limits on the first complex datasets.
Diving into the World of Optimization
This hackathon gave me the opportunity to dive deep into a field I've been passionate about for a long time: applied mathematics for optimization.
Over the weeks, I discovered and used several industry-standard solvers:
- Gurobi — Industry-leading commercial solver for mathematical optimization
- IBM CPLEX — IBM's industrial optimization suite
- HiGHS — High-performance open source solver
- CP-SAT — Google OR-Tools constraint solver
But beyond the tools, what struck me most was the underlying logic: how a real-world problem can be transformed into an exploitable mathematical model.
Algorithmic Approaches Explored
I progressively explored different algorithm families:
Metaheuristics:
- Large Neighbourhood Search (LNS) — Iterative destruction and reconstruction of solutions
- Adaptive Large Neighbourhood Search (ALNS) — Adaptive version that dynamically adjusts operators
- Simulated Annealing — Thermodynamic inspiration to escape local optima
- Genetic Algorithms — Evolution of solution populations
Exact Optimization:
- Mixed-Integer Programming (MIP) — Integer programming
- Mixed Integer Linear Programming (MILP) — Linear variant of MIP
Each method brought its strengths, limitations, and most importantly, a different way of reasoning about the problem.
What This Hackathon Really Taught Me
Beyond the code, this experience taught me one essential thing:
Performance doesn't come solely from the algorithm, but from how the problem is formulated.
Today, we almost all have access to the same tools. Solvers are accessible, open source libraries are numerous, and language models are widely democratized.
Technical resources are therefore becoming standardized.
The difference is no longer just about computing power or the tool used, but about:
- Deep understanding of the problem
- The ability to reformulate it correctly
- The assumptions we choose to make
- How we interact with these tools to guide them in the right direction
Knowing how to break down a complex problem, isolate the right levers, structure a search, or guide a solver often becomes far more decisive than the apparent sophistication of the algorithm itself.
Optimization: An Omnipresent Yet Invisible Field
What makes this field particularly fascinating is its ubiquity.
Behind sometimes abstract mathematical models lie systems used every day, often without us even realizing it:
- Telecom network planning
- Delivery route optimization (Vehicle Routing Problem)
- Cloud resource allocation
- Energy management (Unit Commitment Problem)
- Air and rail transportation
- Industrial scheduling
- Dynamic pricing
All systems where a few percentage points of improvement can represent considerable savings at industrial scale.
A Foundational Experience
Finishing 1st out of nearly 850 participants in the qualifying rounds was obviously a great personal satisfaction, but above all a confirmation of a deep interest in problems combining mathematics, data, modeling, and complex systems.
This hackathon allowed me to consolidate a way of thinking, approach problems differently, and better understand what really drives performance in demanding technical environments.
It's precisely this type of subject — at the frontier between theory and real-world impact — that I want to continue exploring and deepening.
To discover my other projects, check out my projects page.