Multi-Agent AI Simulation – Bullwhip Effect

Built a Python-orchestrated simulation with three agent types: an LLM agent (Mistral-7B), a human-modelled agent using behavioural heuristics, and a rule-based robotic agent; ran experiments across multiple demand uncertainty scenarios measuring decision variance, failure modes, and escalation patterns. Delivered an Agentic AI PoC with structured quantitative comparisons and stakeholder-ready reports demonstrating relative strengths and failure conditions of each agent type.

Supply chain demand amplification (the Bullwhip Effect) causes costly inventory mismatches, and it was unclear how AI-based decision-making compares to human and rule-based approaches under uncertainty.

Python-orchestrated simulation with three agent types (Mistral-7B LLM, human-modelled behavioural heuristics, rule-based robotic) across multiple demand scenarios with quantitative comparisons.