Operations Research·Stochastic Optimization·Scientific Software

Ahmed Alqershi

I build clean things.

Based in Türkiye·Open to relocation

01About

Building a model is one job; building the software it lives inside is another. I work across both.

My main work is in mathematical optimization and scientific software. At GAMS Software GmbH, I am a core developer of GAMSPy and the primary implementer of its SDDP framework for multistage stochastic programs. I also work on economic modelling and applied simulation–optimization projects.

I studied Industrial Engineering before completing a thesis-based M.Sc. in Electrical and Computer Engineering. That route gave me an operations-research foundation together with experience in computational modelling, experimentation and software development. My current research interests include stochastic decomposition, multiscale energy-system planning and the computational choices that decide whether large models can actually be solved.

Everything below is real work — the research, the projects, the experience, even the puzzle a few screens down, which is a genuine optimization solver rather than a decoration.

Stochastic decompositionEnergy-system modellingScientific software

02Experience

Research, consulting, and the factory floor — every role has circled the same question: how do you make a system run better?

I work(ed) with

GAMS Software GmbH

Operations Research Analyst

January 2023 – Present

Remote

Operations Research Analyst

Erasmus Trainee·September – December 2022

Braunschweig, Germany

Researched multi-objective optimization and implemented an initial executable example based on the sandwich algorithm, together with technical presentations and documentation.

I work on GAMSPy, GAMS's Python modelling library. I am the core developer and primary implementer of its SDDP framework for multistage stochastic programs. I have also contributed neural-network and ReLU formulations, bound propagation and piecewise-linear APIs, the model library, Connect data-integration work, tests, documentation and bug fixes.

03Education

An operations-research foundation, then the computational side of it — both degrees taught and examined in English.

2022

Industrial Engineering

B.Sc.

Abdullah Gül UniversityAugust 2018 – June 2022

245 ECTS
Taught and examined entirely in English

Completed 57 ECTS in operations research and optimization, including Stochastic Models, Deterministic Optimization, Mathematical Modeling, System Simulation and Decision and Risk Analysis; also completed 32 ECTS of supervised industrial work.

GPA 3.84 / 4.00High HonourFirst-ranked graduate, 2022

2025

Electrical and Computer Engineering

M.Sc.

Abdullah Gül UniversitySeptember 2023 – May 2025

120 ECTS
Thesis-based degree, taught and examined entirely in English

Thesis (45 ECTS): Neural Insights into Drug Repositioning: A Literature-Based Framework Using Word Embeddings and Siamese Networks

Designed a TensorFlow/Keras pipeline combining SemMedDB relations, FastText embeddings and Siamese neural networks. Evaluated 576 experimental configurations; the selected model reached 87.66% validation accuracy and 83.2% test accuracy.

GPA 3.96 / 4.00

04Research

A manuscript under review, a published technical article, a thesis, and the research software I am responsible for at work.

Journal manuscriptUnder review
A Siamese Neural Network Framework for Computational Drug Repositioning Using Literature-Derived Biomedical Knowledge
Ahmed Alqershi and Gokhan Bakal
Revised manuscript under review at the Journal of Computational Science
Role: First author
Developed from my M.Sc. thesis. I designed the original study, implemented the TensorFlow/Keras pipeline, ran the 576 experimental configurations, analysed the results and wrote the initial manuscript. My co-author and supervisor subsequently helped develop the work into the revised journal manuscript and led substantial parts of the response to peer review.
Technical article
Speed vs. Guarantees: A Practical MIP–NLP Trade-off for NN Robustness
Ahmed Alqershi, Burak Usul, Michael Bussieck, Muhammet Soytürk and Steve Dirkse
GAMS Technical Blog · 12 November 2025
Role: Lead author
Benchmarked exact big-M MIP and complementarity-based NLP formulations for neural-network robustness across 30 MNIST architectures, then tested Sobol multi-starts as a practical way to improve the faster NLP approach.
Read the article
Research software
SDDP framework for GAMSPy
GAMS Software GmbH · released in GAMSPy 1.25.0
Role: Core developer and primary implementer
A reusable framework for multistage stochastic programs with sampled forward passes, backward scenario solves, LP-dual-based value-function cuts, bound tracking and convergence checks. I also developed GAMSPy's GUSS support to batch related scenario solves after profiling showed that repeated Python–GAMS round trips were the main bottleneck.
Documentation
Master's thesis
Neural Insights into Drug Repositioning: A Literature-Based Framework Using Word Embeddings and Siamese Networks
M.Sc. thesis · Abdullah Gül University · 2025 · 45 ECTS
Role: Sole author
A literature-based framework for drug repositioning: SemMedDB relations and FastText embeddings feeding a Siamese network that scores how likely a drug treats a given disease, evaluated across 576 experimental configurations.
YÖK National Thesis Centre
Council of Higher Education (YÖK) National Thesis Centre, thesis no. 958165 — the record is retrieved by searching that number.

Teaching

Volunteer Teaching Assistant — Stochastic Models

Abdullah Gül University·October 2021 – January 2022

After earning an A in the course, I volunteered to help the following cohort. I joined practical sessions, worked through numerical examples, answered questions outside class and recorded public tutorials on Markov chains and Poisson processes.

Watch tutorials

05Skills

Stochastic Optimization & Decomposition

Decisions made now against a branching, uncertain future.

SDDP
Benders decomposition
Multistage stochastic programming
LP / MILP / NLP

Mathematical Modelling & Scientific Software

The models, and the libraries other people build models with.

GAMS
GAMSPy
Python
Model/solver interfaces
API design
Unit testing
Profiling
Documentation

GAMS / GAMSPy: professional modelling and development experience.

Optimization Solvers

What the formulations are actually handed to.

CPLEX
Gurobi
HiGHS
CONOPT

Simulation & Applied Modelling

Where the model has to meet a real plant or a real economy.

Simio
OptQuest
Discrete-event simulation
Simulation–optimization
CGE modelling

Programming & Data

The languages and tooling the research work runs on.

Python
Julia
SQL / PostgreSQL
C#
Java
Git
Docker

Python: primary language, used professionally for scientific-library and API development. Julia: working knowledge — I studied SDDP.jl's API and architecture while designing GAMSPy's SDDP interface.

Machine Learning

From the thesis, and from ML formulations inside GAMSPy.

TensorFlow/Keras
PyTorch
scikit-learn
FastText
Siamese neural networks

Web Development

Supporting skill — how a model reaches the people using it.

TypeScript
React
Next.js
Django
PostgreSQL
Tailwind

Supports building analyst-facing platforms. Not my primary research identity.

Languages

Arabic — Native
English — Fluent
Turkish — Good

Ordered by how central each one is to my work, not by how long the list is. The first four are where I spend my time.

06Projects

Research software, applied operations research on real production lines, and the platforms built around models other people have to use.

Research software · GAMS Software GmbH

SDDP Framework for GAMSPy

The starting point was a multistage hydrothermal power-planning example based on Vattenfall Energy Trading data, reproduced first and then generalized into a reusable SDDP framework for GAMSPy: sampled forward passes, backward scenario solves, LP-dual-based value-function cuts, bound tracking, convergence checks, and an interface for user-defined stages, states and discrete uncertainty. When the first general version proved too slow, profiling put the bottleneck in repeated Python–GAMS round trips, so I added GUSS support to batch related scenario solves. Released in GAMSPy 1.25.0 with tests, examples and documentation.

SDDP
Stochastic Programming
GAMSPy
Python
Decomposition

Platform · Kaizen Consulting

Dynamic CGE Modelling Platform

A reusable dynamic computable general equilibrium platform — the economic models that simulate how an entire economy responds to a policy change or shock, sector by sector. It covers formulation, calibration, solver integration, data workflows and the analyst-facing software, and was validated on a 10-year Saudi-economy case with 20 sectors and 84 commodities. I was project manager and lead developer, coordinating a six-person team of three senior economists and three developers.

CGE Modelling
Economic Modelling
Calibration
Platform
Team Lead

Applied OR · Stryker Corporation

Hybrid Simulation–Optimization

A plant-wide value-stream map identified the plastic-injection station as the main source of lead time. We built a Simio discrete-event model covering 90 parts, 72 molds and three machines, and used OptQuest to tune a pull-based (Q,R) policy under a 90% service-level requirement. My main contribution was integrating Simio with a Python/Gurobi production scheduler through C# and a database. Year-long 20-ECTS capstone.

Simio
OptQuest
Discrete-event Simulation
Gurobi
Simulation–Optimization

Applied OR · HES Kablo

Production Scheduling

Full-line simulation identified three machines at approximately 98% utilization; the resulting flow shop was then formulated as a makespan-minimizing MILP. I developed a sub-second heuristic extending Johnson's rule — across 30 demand samples it matched the proven optimum 19 times, with a mean deviation of 2.6 minutes — and helped deliver the scheduling application. The plant study estimated a 32% increase in effective production capacity.

MILP
Flow Shop
Scheduling Heuristics
Simulation
Python

M.Sc. Thesis · Abdullah Gül University

Drug Repositioning with Siamese Neural Networks

Drug repositioning looks for new therapeutic uses of already-approved drugs — far faster and cheaper than developing one from scratch. My thesis framed it as a similarity problem: FastText embeddings over biomedical literature relations (SemMedDB) and known drug–disease pairs (RepoDB), fed into a Siamese neural network that scores how likely a drug treats a given disease. Built in TensorFlow/Keras and evaluated across 576 experimental configurations.

TensorFlow/Keras
Siamese Networks
FastText
Biomedical NLP
Drug Repositioning

Platform · Kaizen Consulting

Tourism Satellite Account Toolkit

A toolkit built for a national Ministry of Tourism to compile its Tourism Satellite Account — the international standard for measuring what tourism contributes to an economy. It assembles the TSA tables, validates them for consistency, and publishes the resulting indicators with full traceability.

Tourism Satellite Account
Data Validation
Full-stack
Government

Platform · Kaizen Consulting

Integrated Pricing Platform

A platform for the firm's pricing committee — the team that prices project bids. Instead of ad-hoc spreadsheets, they enter the project's scope, the resources it needs, discount rules and more; the platform structures the pricing decision and generates the final bid report, end to end.

Internal Tooling
Full-stack
Reporting
Platform

Two more pieces of software are live on this page — the site itself, and the Detour game just below, whose solver is exact. Both were built from scratch and are open for you to poke at.

07Detour

The Traveling Salesman Problem asks a deceptively simple question: given a set of cities, what is the shortest loop that visits every one and returns home? It is one of the classic problems in operations research — easy to state, famously hard to solve, and the seed of a whole family of optimization algorithms.

This is its prize-collecting cousin — what optimization calls the Orienteering Problem. Every city is worth points, but your travel budget won't reach them all. Pick the most valuable loop you can afford, then see how close you got to the optimal haul.

Loading today's puzzle…

08Contact

Or just email me at alqershiahmed20@gmail.com.

Ahmed Alqershi

GAMS Software GmbH

Industrial Engineering

Electrical and Computer Engineering

A Siamese Neural Network Framework for Computational Drug Repositioning Using Literature-Derived Biomedical Knowledge

Speed vs. Guarantees: A Practical MIP–NLP Trade-off for NN Robustness

SDDP framework for GAMSPy

Neural Insights into Drug Repositioning: A Literature-Based Framework Using Word Embeddings and Siamese Networks

Volunteer Teaching Assistant — Stochastic Models

Stochastic Optimization & Decomposition

Mathematical Modelling & Scientific Software

Optimization Solvers

Simulation & Applied Modelling

Programming & Data

Machine Learning

Web Development

SDDP Framework for GAMSPy

Dynamic CGE Modelling Platform

Hybrid Simulation–Optimization

Production Scheduling

Drug Repositioning with Siamese Neural Networks

Tourism Satellite Account Toolkit

Integrated Pricing Platform