SENA
SAYGIN ŞENYÜZ
INDUSTRIAL ENGINEER · ARTIFICIAL INTELLIGENCE · BUSINESS ANALYSIS
BETWEEN TÜRKİYE AND SERBIA · FULLY ALIGNED WITH EUROPEAN HOURS · REMOTE
I planned production on the factory floor; now I solve the same problems with machine learning. I lived the delays on the line — now I see them coming.
- M-012021–2024ON THE FACTORY FLOOR
- M-02180,519ORDER RECORDS ANALYSED
- M-0333NODES · AI AGENT
- M-040.39 → 0.71F1 · BASELINE → TUNED
OP-10LINE CONVERSIONmaterial flow → data flowCOMPLETE
← scroll the diagram sideways · arrow keys also work →
The line that moves material in a factory and the system that moves data here are the same object — same stations, same buffers, same scrap branch. Only what flows has changed.
OP-20CONTROL CHARTcontrol limit = decision thresholdLIVE
← scroll the chart sideways · arrow keys also work →
- true alarm
- false alarm
- missed delay
- correctly passed
Measured threshold table — delays caught / alarm precision: 0.35 → 99.8% / 57.4% · 0.40 → 78.0% / 66.4% · 0.50 → 56.0% / 87.4% · 0.55 → 54.4% / 88.7%. Class balance: 54.8% late, 45.2% on time.
The sample is representative; the ratios are the project's measured values. The right limit is a question of cost, not of mathematics — the decision belongs to the business, not the model.
Dataset: DataCo Smart Supply Chain (open data, 180,519 order records) · code and measurements: github.com/senasayginsenyuz/supply-chain-late-delivery-ml ↗
OP-30PROCESS ANALYSISfour layers, one person
BOM, ERP, bottleneck — I saw all of it first-hand. Not in a meeting room.
Cleaning, joining. The stage that is won — or lost — before a model is ever built.
The skill is not racing four models; it is picking the one threshold aligned with business cost.
Who does what differently, and when? Without an answer, the analysis is not finished.
OP-40WORK ORDERSprojects
An end-to-end ML pipeline over 180,519 orders. Three columns that are only filled in after delivery (Delivery Status, Days for shipping (real), shipping date) were detected and removed, taking the feature set from 53 to 25. With the leakage gone the baseline F1 is 0.39; hyperparameter and threshold tuning takes it to 0.71. Train 0.713 / test 0.714 — no overfitting. The threshold was measured at four points (0.35–0.55). Data ceiling: added features left F1 unchanged — the true drivers (weather, traffic, carrier reliability) are absent from the dataset.
COMPLETE WO-2602 Document Control AgentN8N · GEMINI · NOTION · TELEGRAMA 33-node agentic workflow: classifies incoming mail by urgency and document type, archives attachments, and answers questions by querying Notion as a tool. A memory-backed LLM agent with a command interface.
COMPLETE WO-2603 Linear Regression From ScratchNUMPY · GRADIENT DESCENTCost function, gradient and the update rule written by hand — no library shortcut.
COMPLETEMathematical optimisation of the production sequence under constrained machines and labour.
QUEUED · Q4 2026OP-50ROUTE HISTORY2016 → today
- End-to-end ML project and a 33-node AI agent — both published
- Machine Learning Specialization — Stanford & DeepLearning.AI
- ECBA® business analysis certification — IIBA
- Data Analysis School (YÖK) — two AI modules, ~85 h each: generative AI and large language models; supervised and unsupervised learning; tooling that speeds up data analysis
- Python for Everybody (PY4E) — University of Michigan
- Pulled material shortages from the BOM and managed the production calendar
- Procurement coordination; order tracking across four project-based jobs
- Ran two planning projects end to end; 15+ Gantt charts in MS Project
- Order tracking, material planning and stock control through Logo ERP and R-MES
OP-60SKILLS INVENTORYthree disciplines, one bench
LANGUAGES — Turkish (native) · English (B2)
OP-70CERTIFICATES & TRAININGeight certificates · 2023 → 2026
DISPDISPATCHcontact
If there is a problem in your production or supply chain waiting to be solved with data — let's talk.
ROLES I'M TARGETINGAI / data analyst for manufacturing and supply chain · AI business analyst · agentic AI and process automation
MESSAGE FORMThe form opens with JavaScript; if it is off, you can reach me on LinkedIn — link opposite.
LOCATION & WORKING MODEBetween Türkiye and Serbia · fully aligned with European hours · remote