Projects

Each project shows its current status next to the title. No product codes or real production data from the internship are shared.

Internship work

Carpet Production Platform: planning, efficiency, lean

Working, awaiting installation
Where
Karmen Halı, Gaziantep. Production planning internship, September – December 2026
Scope
3 modules on a shared database; Weaving and Finishing departments
Tools
Python, Streamlit, Excel

Problem

Production data had been entered into monthly Excel forms since 2017. The same data was typed twice, once into the production file and once into the OEE files. Planning, efficiency tracking and lean records lived apart from each other.

Solution

I combined three separate tools into one application through a shared integration contract. The modules use the same database, so a change in OEE in the efficiency module can be followed alongside the lean techniques being applied.

Current state

As of November 2026 the three modules run as one application. It currently runs in the cloud (Google Colab); installation on a single central computer on the factory network is planned. I am not reporting a measured business result yet.

Working
Module 1
Production planning
Input / output
Excel order list → plan as Excel, PDF, HTML

Carpet weaving production plan optimisation

  • I developed an optimisation tool (Python) that produces a weaving production plan from order groups.
  • It handles rectangular and oval products and four edge types (oval fringe, overlock, fringe, glued) separately.
  • I put it behind an interface (Streamlit) where the planning team uploads an Excel file and downloads the plan without seeing any code.
  • Scope limit: the tool produces the plan but does not assign it to looms. Loom capability is kept as an optional field to be filled in later.
In trial
Module 2
Production and efficiency
Scope
14 weaving looms; Excel history kept since 2017

Production tracking and OEE (Overall Equipment Effectiveness) system

  • I designed daily production entry per shift and per loom for 14 looms; a single entry replaces double data entry.
  • It records downtime reasons separately (weft search, jute, pile, warp, other); short stops are entered as counts and long stops in minutes.
  • It calculates OEE and produces daily, weekly, monthly and yearly reports.
  • Loom status and work orders with their programme numbers are followed on the same screen.
  • It is designed to import the monthly Excel forms kept since 2017.
In development
Module 3
Lean manufacturing
Scope
Administration, Weaving, Finishing and their sub-departments

Lean manufacturing tracking and reporting module

  • I designed 5S audit, Kaizen log, action tracking and takt/capacity modules with company-wide and per-department views.
  • I prepared a management report and a monthly report template from the Finishing department's OEE spreadsheet; the OEE target is 85%.
  • Reports come out as two kinds of PDF: a presentation-style summary and a detailed report.
  • Design decision: I moved the OEE calculation out of this module and left it to the efficiency module, so the same calculation is not kept in two places.
Completed
Area
Standardisation and training

Standardising defect codes

The factory had no defined defect code system before. I standardised the defect codes and trained the staff on them.

Analytics and modelling

Graduation project
Method
LSTM demand forecasting + integer linear programming (ILP)

Smart production planning and decision support system

A hybrid model that builds a production plan under limited resources. In the model results, operational cost fell by 10–15% and the service level reached 95%.

Personal project
Data
NASA aircraft engine sensor data (Kaggle)
Tools
Python, Pandas, Scikit-learn

Predictive maintenance: remaining useful life estimation

I built a regression model that estimates remaining useful life (RUL) from engine sensor readings and operating settings. I merged the data, engineered features and evaluated performance with MSE. The aim is a better maintenance schedule and less unplanned downtime.

Early stage
Tools
Python, SimPy

Discrete-event simulation of a production line

A learning project that simulates a production line for bottleneck analysis and "what if" capacity scenarios. It belongs to the same family of methods as commercial packages such as Arena and FlexSim.

Early stage
Topic
Supply chain, game theory

Beer game simulation for the carpet supply chain

A simulation that adapts the beer game to the carpet industry supply chain: inventory policy, lead time and the bullwhip effect. It is the first of the game theory studies I plan for the carpet sector.

Proposal stage
Programme
TÜBİTAK 2209-A
Duration
6 months

Simulation optimisation for garment line productivity

A research proposal that studies line productivity with an open data set (UCI garment worker productivity) and combines SimPy simulation with Bayesian optimisation. No company data is used.

Research and community projects

Research project
Call
HORIZON-CL5-2024-D4-01-01

Horizon Europe project application

For the call, I researched potential project partners and suppliers in Europe, carried out literature and market research on applicable technologies and solutions, and contributed to developing the application content.

Research project
Partnership
Türkiye – United Kingdom

AI4Waste: construction and demolition waste management with AI

I carried out literature and technology research on real-time detection, classification and reuse of construction and demolition waste with AI, and supported interdisciplinary project development between stakeholders in the two countries.

Completed
Funded by
Ministry of Youth and Sports, ÜNİDES programme

Reading the Data: A Step into the Digital World

As president of the Gaziantep University Software Society I wrote the application and managed the project: digital literacy, data science and machine learning training for university students.

Automation with AI

Planned
Purpose
Learning

Multi-agent workflows

  • I am learning to set up a workflow on cloud servers in which different AI models share coding, document preparation and market analysis tasks.
  • Design of a notice tracking system: one agent watches public notices, one extracts the requirements, and one matches and notifies suitable people.