Ali Rouhani | OR Scientist • Data Scientist • Data Engineer • Applied AI

Montreal, Canada • ali.rouhani@umontreal.ca

Ali Rouhani

Operations Research | Machine Learning | Data Engineering | Supply Chain Planning

I build high-performance software engines that solve complex logistics, data, and scheduling problems. My work connects rigorous mathematical optimization, machine learning, and robust data engineering to turn massive operational uncertainty into predictable cost reductions. As a Ph.D. candidate at the Université de Montréal and researcher at CIRRELT, I design scalable systems that simultaneously coordinate millions of dollars in moving assets. Whether you are looking to accelerate slow-running optimization solvers, deploy automated forecasting pipelines, or structure chaotic supply chain data into high-throughput cloud databases, I build production-ready tools that scale.

LinkedIn GitHub Scholar

Focus areas

Optimization: Accelerated Benders decomposition, Matheuristics, Stochastic programming

Data Engineering: ETL pipelines, PostgreSQL, Cloud Deployments (Azure, Docker)

Applied AI & ML: Hierarchical clustering, Ensemble forecasting, LHS scenario generation

Supply Chain Planning: Multi-echelon network design, freight routing, capacity planning

Keywords

Accelerated Benders LightGBM / XGBoost ETL & SQL Python/C++/Delphi Hierarchical Clustering Decision Support

About My Value Proposition

Operations Research & Logistics

I design smart scheduling systems that coordinate trucks, drivers, and freight simultaneously. My custom exact solution frameworks (accelerated Benders decomposition) and Matheuristics bypass standard solver limits, cutting baseline operational costs by over 55% and boosting execution speeds by 60% in HPC environments.

Data Science & Applied AI

I replace risky, fixed-demand assumptions with flexible, multi-scenario planning frameworks and hierarchical data clustering to completely eliminate stockouts. I program high-granularity, multi-model ensemble pipelines to capture complex time-series trends and validate probabilities using rigorous LHS sampling.

Data Engineering & Infrastructure

I architect centralized cloud databases, build automated data-cleaning pipelines, and containerize software tools using Azure, PostgreSQL, and Docker. I develop hybrid architectures that bind high-speed C++ pathfinding algorithms to clean Python user interfaces to ensure reliable, high-speed production deployments.

Core Skills & Technical Arsenal

Supply Chain & Planning

Demand Forecasting Inventory Optimization Network Design Capacity Allocation Stockout Prevention S&OP Support

Optimization / OR

Accelerated Benders Dantzig-Wolfe Matheuristics Stochastic Programming Lagrangian Relaxation

Data Science & ML

Hierarchical Clustering LHS Sampling Time-Series Forecasting XGBoost & LightGBM Probability Validation

Software Engineering

Python (pybind11) C++ (C++17/20, OpenMP) Delphi SQL / PostgreSQL Docker & Azure CI/CD

Solvers & Tooling

Gurobi & CPLEX OR-Tools Scikit-Learn Pandas & NumPy Pytest & GoogleTest

Professional Experience

Supply Chain Optimization & Analytics Researcher Jan 2021 – Present

CIRRELT, Université de Montréal • Montreal, Canada

Logistics Network Design: Formulated a synchronized supply chain planning system that simultaneously schedules interdependent transportation assets (like matching trucks with drivers or cargo blocks with trains) to optimize worldwide and regional freight operations.
Cost & Route Optimization: Evaluated deterministic point-estimate demand profiles, demonstrating that transitioning to this synchronized network layout yields over 55% in operational cost savings against standard direct-routing models across 500,000+ transit arcs.
Risk Mitigation & AI Clustering: Replaced risky fixed-demand planning with a flexible multi-scenario model. Applied hierarchical clustering algorithms to group and simplify massive datasets of joint capacity and demand uncertainty, completely eliminating stockouts across volatile networks.
High-Performance Solver Engineering: Built the core mathematical solver engine using accelerated Benders decomposition and high-performance parallel computing, creating efficient memory structures to bypass standard monolithic scalability limits.
Performance Optimization: Accelerated overall model execution speeds by 60% inside high-performance computing (HPC) environments, and implemented strict unit and integration testing pipelines to guarantee reproducibility across model revisions].
Stakeholder Communication: Presented complex logistics models and predictive analytics frameworks at international conferences (JOPT & ISMP), translating heavy mathematical strategies into clear, actionable business insights for diverse corporate planners.

Data Analyst & Operations Researcher (M.Sc.) Feb 2018 – Dec 2020

Shahed University

Collaborative Logistics: Formulated a multi-stage planning framework for a collaborative distribution platform that allows warehouses and delivery locations to share inventory and flexible capacity during major supply shortages.
LHS Scenario Generation & Validation: Generated realistic disaster and demand scenarios utilizing Latin Hypercube Sampling (LHS) and data clustering algorithms. Proved the reliability of the LHS scenario generation by calculating and comparing the area under the curve of the final Probability Density Function (PDF) against the original 1,000-scenario distribution, mathematically validating zero critical data loss.
Disaster Relief Networks: Designed a stochastic logistics framework to optimize post-disaster humanitarian aid distribution, coordinating the setup of local relief facilities to prevent network congestion under uncertain demand.
Matheuristics & Speedup: Designed custom Matheuristics to solve large-scale logistics and humanitarian problems, cutting network planning execution runtimes by 50% over traditional exact solvers while maintaining an average optimality gap below 0.1%.
Technical Mentorship: Mentored undergraduate and junior master's students, providing technical guidance on research methodologies, algorithm design, and modern software engineering practices.

Data Engineer & Software Developer Jun 2017 – Jun 2018

Afrand Sana’t Company

CRM & Database Design: Designed and developed a centralized customer relationship management (CRM) application and sales management dashboard system using Delphi and SQL Server.
ETL & Data Migration: Built automated data cleaning pipelines to consolidate inconsistent records, migrating fragmented Excel files into a unified relational database to improve organizational data quality.
Query Optimization: Optimized complex database SQL queries and stored procedures, significantly enhancing enterprise reporting generation speeds and overall software responsiveness.
Workflow Automation: Streamlined client tracking and reporting workflows, eliminating data redundancy and reducing manual data handling for operational and administrative teams.

Applied Projects & Core Engines

flowbalance — High-Performance Asset Repositioning Engine • GitHub C++, Python, pybind11, Google OR-Tools, Pandas, Pydantic

Dynamic Network Optimization: Created an open-source universal network flow program to track dynamic states and optimize asset repositioning across time-space networks. Adaptable across multiple domains:
- Logistics & Freight: Nodes are cities, edges are highways, and transit time is driving duration.
- Supply Chain & Inventory: Nodes are warehouses, edges are internal transfers, and holding costs are storage fees.
- Production Planning: Nodes are assembly machines, edges are processing actions, and transit time is machine cycle time.
- Finance & Treasury: Nodes are corporate bank accounts, edges are wire transfers, and transit time is the banking clearing period.
High-Speed Backend: Engineered a fast C++ pricing engine to solve massive Multi-Commodity Network Flow problems using exact Dantzig-Wolfe decomposition, implementing custom priority-queue search algorithms to quickly dynamically generate negative reduced-cost paths.
Hybrid Software Architecture: Built the Python-based Restricted Master Problem (RMP) using Google OR-Tools (GLOP), connecting it directly to the C++ engine via pybind11 integrated with a dual-variable pricing filter that significantly reduces subproblem computations.
Relational Integrity: Designed strict timeline validation schemas using Pydantic to map abstract temporal inventory states flawlessy, and utilized Pandas to automatically translate the mathematical outputs into clean operational reports.
Strategic Support: Expanded the engine's applicability to offer strategic decision support for empty container repositioning, multimodal freight scheduling, and cross-docking optimization.

Store Sales — End-to-End Time Series Forecasting Pipeline • GitHub Python, LightGBM, XGBoost, PostgreSQL, Docker, Azure, CI/CD

Demand Forecasting: Built an automated retail sales forecasting system to predict daily product demand across multiple store locations, allowing inventory teams to proactively prevent stockouts.
Machine Learning Ensembles: Trained a high-granularity system of 33 independent LightGBM and XGBoost models in Python to isolate family-specific purchasing behaviors. Implemented Recursive Autoregressive Inference to utilize short-term lags (1–15 days), significantly reducing temporal blindness compared to static bulk-prediction methods.
Feature Engineering: Constructed advanced feature sets including volatility-weighted moving averages, exponentially weighted averages (EWMA), and deterministic filters for localized economic cycles.
Data Engineering: Architected a scalable data engineering pipeline leveraging PostgreSQL and Azure Cloud to manage high-dimensional transactional datasets, ensuring data integrity for large-scale ingestion.
Cloud MLOps Stack: Containerized the full MLOps stack using Docker on Azure Virtual Machines, establishing a Continuous Integration (CI) framework for automated training and evaluation.
Performance Impact: Achieved a highly competitive out-of-sample RMSLE score of 0.41036, approaching the elite benchmark of 0.37687 within a zero-inflated, highly complex retail prediction environment.

Demand Analysis Engine — Modular Forecasting & Diagnostic Tool • GitHub Python, LassoCV, Scikit-Learn, Pytest

Demand Profiling: Architected a modular, reusable time-series pipeline for automated schema ingestion, deep demand profiling, and high-dimensional feature engineering to safeguard downstream shipping schedules.
Feature Matrix Construction: Engineered an automated temporal feature matrix in Python extracting high-order rolling windows, multi-period lags, and deterministic calendar indicators to capture complex seasonality and non-linear patterns.
Parallel Model Evaluation: Implemented a parallel multi-model evaluation framework comparing structural regularization engines (LassoCV, OLS) against non-linear gradient-boosted ensembles (LightGBM, XGBoost).
Risk Diagnostics: Designed a specialized out-of-sample diagnostic module to isolate prediction errors, calculating Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Systemic Bias to track operational risks.
Defensive Testing: Established strict defensive pipeline testing routines using Pytest to validate matrix density, enforce shape constraints, and mathematically eliminate data leakage prior to production training phases.
Performance Output: Isolated strong structural linearity in scaled supply chain matrices, achieving an out-of-sample R² score of 0.461 with optimized Lasso regularization while maintaining stable generalization metrics.

Vehicle Routing Research Platform (ML + OR) C++, Python, GoogleTest, GitHub Actions

Routing Optimization: Developed a modular research platform to evaluate advanced optimization strategies and learning algorithms for the Vehicle Routing Problem using Solomon benchmarks.
Algorithm Engineering: Redefined and customized Dijkstra's shortest-path algorithm and greedy heuristics specifically for the VRP. Programmed the core algorithms in modern C++ with an extensible architecture strictly designed for seamless metaheuristic and machine learning integration.
Learning-Assisted Routing: Investigated route-cost prediction, instance feature extraction, and heuristic selection methods to reduce runtime while preserving high solution quality.
Reliable Infrastructure: Integrated rigorous automated unit testing using GoogleTest and CI/CD workflows via GitHub Actions with memory sanitizers to guarantee algorithmic reliability.

Predictive Modeling & Classification Pipelines (Data Science) Python, LightGBM, XGBoost, scikit-learn, SQL

Built robust data extraction and preprocessing pipelines for customer behavior prediction (Bank Churn) and demographic survival analysis (Titanic dataset).
Conducted extensive exploratory data analysis and engineered numerical and categorical features to improve model generalization.
Systematically tuned hyperparameters for LightGBM, XGBoost, and Logistic Regression models to maximize predictive accuracy across varying dataset distributions.

Publications & Research

Benders Decomposition for Two-Layer Network Design

Publication: Working Paper, INFORMS Journal on Computing
Industry translation: Scalable network investment and capacity planning; faster planning cycles via decomposition.
Methods: Accelerated Benders decomposition, Valid inequalities, Matheuristics → C++.

Two-Layer Network Design under Capacity Uncertainty

Publication: Working Paper, European Journal of Operational Research
Industry translation: Risk-aware capacity decisions via stress-tested network design across scenarios.
Methods: Benders decomposition, Hierarchical clustering, Branch-and-Cut → C++.

Stochastic Two-Layer Network Design under Joint Uncertainty

Publication: Working Paper, Computers & Operations Research
Industry translation: Advanced modeling for two-layer networks accounting for bidirectional uncertainty propagation.
Methods: Accelerated Benders decomposition, Scenario clustering → C++.

Dynamic capacity (re)-allocation in collaborative distribution platforms under disruption uncertainty

Industry translation: Operational re-planning under disruptions; dynamic capacity allocation when conditions change.
Methods: Multi-stage stochastic programming, LHS Scenario generation, Matheuristics → Julia.

Accelerated Benders Decomposition for Stochastic Liner Shipping Fleet Repositioning

Industry translation: Asset rebalancing under uncertainty to improve cost and service reliability in maritime logistics.
Methods: Two-stage stochastic programming, Benders decomposition, Valid inequalities → Julia.

A Lagrangian Decomposition Algorithm for Robust Green Transportation Location Problem

Publication: International Journal of Engineering (2019).
Methods: Lagrangian decomposition.

Education & Awards

Education

Ph.D. in Computer Science (Operations Research), Université de Montréal — 2021 – Present (Expected 2026)
M.Sc. in Industrial Engineering (System Optimization), Shahed University — 2016 – 2019
B.Sc. in Industrial Engineering, University of Qom — 2011 – 2016

Presentations & Awards

Conference Presentations: JOPT, ISMP 2024
UdeM Exemption Scholarship for International Students (2021)
Best Undergraduate Researcher, University of Qom (2016)
3rd Place, Tehran Mathematics Olympiad (2006)