Impact
Built models where local agent rules produce measurable system behavior.
Impact
Compared scenarios through repeatable simulation runs.
Impact
Explained emergent outcomes with charts, parameters, and sensitivity notes.
Deliverables
- Simulation model
- Scenario experiment set
- Results interpretation
Technical Overview
Systems such as epidemiological spread, evacuation, markets, and biological populations cannot always be understood from aggregate formulas alone. The project needed models where individual rules, constraints, and randomness could be inspected directly.
Architecture
flowchart LR
initialize_agents[Initialize agents] --> local_rules[Apply local rules]
local_rules --> interactions[Move and interact]
interactions --> metrics[Collect metrics]
metrics --> clock_check{More ticks}
clock_check -->|yes| local_rules
clock_check -->|no| scenario_compare[Compare scenarios]
Model
as an aggregate comparison point; the agent model implements infection and recovery at the individual level rather than only through compartment counts.
Implementation Sketch
to go
ask agents [
move
interact-with-neighbors
update-state
]
collect-metrics
tick
endEngineering Approach
- Define agents, state variables, environment constraints, and update rules.
- Run parameter sweeps to test whether outcomes are stable or highly sensitive.
- Separate model behavior from real-world claims by documenting assumptions and calibration limits.
Results
- Built agent-based simulations in NetLogo and Python for outbreak dynamics, movement, competition, and social behavior.
- Used repeated runs to distinguish stable patterns from random artifacts.
- Documented parameter effects so the model could be adjusted without rewriting the core rules.
What This Demonstrates
- Simulation projects need explicit assumptions because model realism is never automatic.
- Sensitivity analysis is often more informative than a single run.
- Good visual output helps validate whether the model is behaving as intended.