From raw data to shipped systems.

I'm Ates, an Industrial Engineering student at TU/e, looking for a summer 2027 internship in software, finance or operations.

See the work Download CV

Modelling is the slowest step, so jobs pile up in front of it.

Each box is a step from raw data to a finished product, and each blue square is a job. Most of a job's time is spent waiting in a queue, not being worked on. Tap a step to improve it; tap again to undo.

Total time per job0.0 s

Building for a real team.

Two stints at Derin Bilgi in Istanbul: a summer internship, then a part-time developer role.

Derin Bilgi · Summer Intern

Jun. 2025 – Aug. 2025

Istanbul, Türkiye (hybrid)

Contributed to application development and debugging alongside experienced developers, then returned in Mar. 2026 as a part-time developer.

Derin Bilgi · Software Developer, part-time

Mar. 2026 – Aug. 2026

Istanbul, Türkiye (remote)

Proposed and built Lup Diye end to end (confidential).
Developed and debugged features in an Agile Scrum team, with weekly sprint planning and progress reviews.

One line of work.

Each project starts as an operations or money question and ends as software someone can use. Pick one and try it here.

Hiring for

EUV Supply Risk

Supply chain

Hold €82m of cheap spares, no optics, and add €59m of equity value.

Every EUV scanner depends on three single-source modules. I simulate 20,000 versions of 2027 to 2030 and price each buffer policy through my ASML DCF.

Simulated futures: 20,000
Recommended buffer: +€59m
On-time delivery: 57% to 63%

ASML EUV · 2027 to 2030

20,000 paths

Buffer policy

Equity value vs no buffer

+€59m

95% CI +€55m to +€62m

2 NXE light sources, 5 NXE + 1 EXE drive lasers, no optics.

Stock held: €82m
On-time delivery: 63%

Recommended buffer, as failures cluster more

+€9m
Independent: +€59m
Central case: +€131m
Doubled

Supplier reliability is not public, so those inputs are labelled illustrative priors with stated ranges. Simulated outcomes, not predictions.

Open the full project Source and README

Python
SciPy
TypeScript
React
Web Worker

How it was built

Problem

Projection optics from Zeiss, light sources from Cymer and drive lasers from Trumpf each come from one supplier. How much stock of each is worth holding?

Data

The Base case of my ASML DCF plus sourced inputs. Supplier reliability is not public, so those inputs are labelled illustrative priors with stated ranges.

Method

48 months per path: industry stress, orders, outages, yields, assembly, late orders and cancellations, with shipments feeding the DCF. Common random numbers keep policy comparisons paired; buffers are optimised on one seed and evaluated on another. 22 tests, and the deterministic case reproduces the DCF price.

Result

Two spare sets of everything destroys €193m, because a late machine is mostly a delayed sale. Supply-chain risk costs about €2.44 a share, so it cannot explain the gap between my €935 and the €1,611 price. The decision depends most on laser and light-source outage rates and on how strongly failures cluster.

The EUV Supply Risk site: the answer is to buffer cheap, failure-prone modules and hold no spare optics, worth +€59m of equity value, while two spare sets of everything lose €193m. — The published study, with an in-browser simulator.

ASML DCF Valuation

Finance

My model says €935 a share. The market says €1,611.

A ten-year discounted cash flow valuation of ASML, built bottom-up from its filings, checked against peers, and published as a model you can change.

Base case: €935
Below market: 42%
Exit multiple the price needs: 33x

ASML Holding N.V. · 7 Oct 2026

market €1,611

Scenario

Terminal value

WACC9.38%Terminal growth2.5%Exit EV/EBITDA on FY3520x

Implied value per share

€935

−42.0% vs market

Reverse DCF: today's price needs a 33x exit multiple, or 7.1% growth forever after 2035.

Same cash flows as the Excel model: FY26 to FY35, mid-year discounting, net cash and investments added to reach equity. Black line: market price.

Open the full project Source and README

Python
Excel
openpyxl
React
TypeScript

How it was built

Problem

ASML trades as the monopoly behind every leading-edge chip. What do its own cash flows say it is worth, and what does today's price assume?

Data

ASML's 20-F and quarterly filings (US GAAP, FY2021 to Q2 2026), market data on 7 Oct 2026, Damodaran's equity risk premium and industry betas, and four peers: Applied Materials, Lam Research, KLA and Tokyo Electron.

Method

Revenue built from systems shipped times selling price for low-NA EUV, High-NA EUV and DUV, capped by ASML's stated capacity; five forecast years plus a five-year fade; WACC of 9.4%; Gordon growth and exit-multiple terminal values; comps, scenarios and a reverse DCF. 27 automated checks tie the Excel model to an independent Python and TypeScript copy.

Result

Base case €935 a share, 42% below the €1,611 price, while peers put it near fair value. Today's price needs a 33x exit multiple or 7.1% growth forever after 2035: the market is pricing the AI cycle as lasting well past a ten-year forecast.

The ASML valuation site: a blended DCF of €935 a share against the €1,611 market price, with a football field comparing the 52-week range, trading comps, Gordon growth, exit multiple and scenario values. — The published valuation site, with the Excel model one click away.

Demand Forecaster

Industrial Eng.

One global model beat the seasonal-naive benchmark by 26% across 500 products.

Forecasts daily demand for 500 store-item series, flags which products run out before the next delivery, and says how much to order.

At risk in 14 days: 178 of 500
XGBoost MAPE: 13.6%
MASE vs naive: 0.74

store 9 · item 1

Runs out on day 13, before delivery

Days until next delivery14

Service level

On hand (simulated): 330
Forecast demand until delivery: 378
Safety stock: 29
Reorder point: 406

Order now77 units

Real backtest: grey is what sold, blue is the XGBoost forecast. Safety stock = z × σ × √days, with σ the daily forecast error.

Open the full project Source and README

Python
XGBoost
SARIMA
Streamlit

How it was built

Problem

A store must reorder before it runs out. Which of 500 products will stock out before the next delivery, and how much should be ordered?

Data

Kaggle Store Item Demand Forecasting Challenge: 10 stores by 50 items, daily sales from 2013 to 2017.

Method

A SARIMA baseline on one series, then one global XGBoost model with lag, rolling and calendar features across all 500. One global model learns the shared weekly pattern and scales; 500 separate SARIMA fits do not.

Result

SARIMA on one series scored MAPE 33.4% and MASE 1.01, no better than repeating last week. XGBoost across all 500 scored MAPE 13.6% and MASE 0.74. 178 of 500 products run out within a 14-day delivery window; each gets a safety stock and reorder point at the chosen service level.

The Demand Forecaster Streamlit dashboard: 500 products tracked, 178 at risk of stockout, a 14-day restock window, and a table of at-risk products. — The Streamlit dashboard.

SEC Filing Analyzer

Finance

Every claim ties to a quote. Every number comes from XBRL, never from the model.

Reads 10-K and 10-Q filings and returns risks, growth bets and priorities, each tied to a quote from the filing itself.

Companies in the demo: 5
Figures written by the model: 0
Revenue history recovered: 4 years

NVIDIA CORP

10-K filed 2026-02-25

Evidence from the filing

“During fiscal year 2026 and fiscal year 2025, we spent $6.1 billion and $3.4 billion on capital expenditures, respectively. We expect to increase capital expenditures in fiscal year 2027 relative to fiscal year 2026 to support the future growth of our business.”

Real output from the app's demo mode. Quotes come straight from each 10-K.

Open the full project Source and README

FastAPI
React 19
TypeScript
Ollama
SEC XBRL

How it was built

Problem

A 10-K takes hours to read, and the change that matters is buried in legal boilerplate.

Data

SEC EDGAR filings (Business, Risk Factors and MD&A sections) and SEC XBRL financial data.

Method

The model reads only the extracted sections and must return strict JSON with a quote behind every claim. Every number is computed in Python from XBRL, never by the model, because language models invent figures.

Result

Five companies open instantly in demo mode; any other filing runs on a free local model. Found and fixed a data bug: NVIDIA changed its XBRL revenue tag after FY2022, which had silently dropped four years of revenue.

The SEC Filing Analyzer on NVIDIA's 10-K: revenue $215.94B, net income $120.07B, 65.5% revenue growth and 55.6% net margin, an executive summary, and three key risks each backed by a quote from the filing. — The app, analyzing NVIDIA's 2026 10-K.

Discount Policy Simulator

Operations

Above 20% off, every category loses money.

Finds where discounts lose money across 51,290 order lines of the public Global Superstore sample dataset, and tests a cap before anyone changes a price.

Order lines: 51,290
Lost above 20% off: −$815k
Profit with a 20% cap: $2.28–2.50M

Problem

Discounts win orders, but which ones cost more than they earn, and what policy would stop it?

Data

Global Superstore, a public sample dataset: 51,290 order lines across 7 markets and 3 categories, 2011 to 2014.

Method

Cleaned in pandas, margin by discount level in Python and SQL, then a what-if model that re-prices orders above a cap from their implied list price and cost, with customer retention as an explicit assumption.

Result

Above 20% off, every category loses money: those orders lost $815k. A 20% cap lifts profit from $1.47M to between $2.28M, if those customers all leave, and $2.50M, if they all stay.

More projects and experiments live on GitHub.

I study how work flows, then write the software that makes it flow better.

Industrial Engineering taught me to see queues, bottlenecks and wasted steps. Finance taught me to read what a company says about itself. Code lets me act on both.

Most of my projects start with a spreadsheet or a filing and end as a tool someone can open in a browser.

Study: B.Sc. Industrial Engineering, TU/e, 2025 to 2028
Languages: Turkish (C2), English (C1), Dutch (A1)
Coursework: Statistics, Data Analytics, Algorithmic Programming, Business Information Systems, Financial and Managerial Accounting
Outside work: Tennis, padel, fitness and following the markets
Based in: Eindhoven, Netherlands

Industrial Engineering

Monte Carlo simulation: EUV Supply Risk
Demand forecasting, stockout risk: Forecaster
Pricing policy analysis: Discount Simulator
Flow analysis: this page

Finance

DCF valuation, WACC: ASML DCF
Trading comps, scenarios: ASML DCF
10-K / 10-Q analysis: SEC Analyzer
Profit and margin: Discount Simulator

Full-stack

React, TypeScript: SEC Analyzer, ASML DCF
FastAPI: SEC Analyzer
Python, pandas, SQL: Discount Simulator
Simulation in the browser: EUV Supply Risk

Pull the next project.

I'm looking for a summer 2027 internship in software, trading/quant, supply chain, operations or investment banking. If you have one, a project or a question, send the signal.

Email meCV