Intelligent-Last-Mile-Logistics-Optimization-System

An end-to-end intelligent logistics optimization system integrating Capacitated Vehicle Routing Problem (CVRP) modeling, Google OR-Tools optimization, LLM-based business analysis, and interactive agent-based scenario replanning.

---

Project Overview

This project simulates a real-world last-mile delivery scenario, where a fleet of vehicles must deliver parcels from a central depot to multiple customers under strict capacity constraints.

The system provides a complete pipeline for data-driven logistics decision-making:

• Optimization-based routing (OR-Tools)
• Baseline heuristic comparison
• KPI-driven performance evaluation
• LLM-powered interpretation of results
• Agent-based interactive replanning via a Streamlit dashboard

---

Mathematical Formulation

The core routing engine optimizes the standard CVRP objective—minimizing total transportation cost while satisfying rigid physical constraints:

Objective Function: $$\text{Minimize} \quad \sum_{k=1}^{K} \sum_{i=0}^{N} \sum_{j=0}^{N} c_{ij} x_{ijk}$$

Capacity Constraint: $$\sum_{i=1}^{N} d_i y_{ik} \le Q_k \quad \forall k \in K$$

---

Methodology

1. Problem Modeling

• Modeled as CVRP (Capacitated Vehicle Routing Problem).
• Single depot + multiple customer nodes.
• Fixed vehicle capacity constraint.

2. Baseline Algorithm

• Greedy nearest-neighbor heuristic.

3. Optimization Solver

• Google OR-Tools routing solver (Guided Local Search).

4. Evaluation Metrics

• Total distance, Total cost, Vehicles used, Average load utilization, Cost per parcel.

5. LLM Copilot

• Converts optimization results into human-readable business insights.

6. Agent Module

• Supports natural language queries for dynamic replanning (e.g., "replan with capacity 120", "compare baseline and ortools").

---

Quick Start

1. Install Dependencies

Clone the repository and install the required packages:

git clone https://github.com/kangwooho0126-prog/Intelligent-Last-Mile-Logistics-Optimization-System.git
cd Intelligent-Last-Mile-Logistics-Optimization-System
pip install -r requirements.txt

2. Launch the Interactive Dashboard (Recommended)

Start the Streamlit web application to access the OR-Tools solver and the LLM Agent:

streamlit run app/streamlit_app.py

3. Run the CLI Optimization Engine

Alternatively, run the core backend solver directly:

python main.py

---

Results & Business Impact

Baseline vs OR-Tools Evaluation

Metric	Baseline (Greedy)	OR-Tools (Optimized)	Improvement
Total Distance	1146.40 km	787.08 km	↓ 31.34%
Total Cost	$3839.20	$2761.25	↓ 28.08%
Vehicles Used	5 Trucks	5 Trucks	=
Load Utilization	82.0%	82.0%	=

Key Insight

"Optimization improves delivery efficiency primarily through better route structuring (reducing overlapping paths and deadhead miles), rather than increasing logistics resources."

---

Agent Replanning Example

The integrated LLM Agent allows dispatchers to dynamically interact with the system using natural language.

User Query:

> "replan with capacity 120"

System Action & Output:

1. Intent Parsing: Agent identifies the intent to change MAX_CAPACITY constraint to 120.
2. Re-Optimization: Triggers the OR-Tools solver seamlessly with the new parameters.
3. LLM Summary: Generates a comparative business report detailing how the reduced capacity impacts fleet utilization and operational costs, displaying the new metrics directly on the Streamlit dashboard.