MMehmet Ünlü
HomeAboutProjectsNotesResumeContact
TR

Mehmet Ünlü

I study Electronics and Communication Engineering at ITU and build hands-on projects around forecasting, computer vision, and making data workflows faster.

HomeAboutProjectsNotesResumeContact

ITU Electronics and Communication Engineering

Mehmet Ünlü

Electronics & Communication Engineering Student

I work on forecasting, computer vision, and practical tooling around data and models through hands-on projects and early product work.

I like problems where the model is only part of the story: messy data, validation decisions, runtime limits, and the small engineering choices that make an experiment usable.

View ProjectsDownload CVGitHubLinkedInContact

Current desk

Istanbul / ITU

Forecasting, computer vision, and making slow data code less painful.

The strongest projects here come from real constraints: many zero-demand rows, drone footage that needs careful motion estimates, and pipelines that must be fast enough to iterate on.

Products

20k

Daily item-level series

History

3 yrs

Rolling temporal windows

Zero demand

85%

Sparse target records

Metrics

WAPE / RMSSE

Forecast quality

Build notes

Keep validation close to the way the project would actually run.
Measure the boring parts: runtime, memory, leakage, and failure cases.
Write case studies so the tradeoffs are visible, not hidden.

What I Work On

I care about models that survive contact with real data.

Most of my work starts with an awkward constraint: too many zeros, slow feature code, noisy motion, or a validation setup that needs to be treated carefully.

Forecasting

Sparse demand, rolling validation, and recursive predictions without peeking into the future.

Validation

Checking whether the setup is fair before trusting the score.

Speed

Replacing slow loops and repeated work with cleaner, faster data transforms.

Pipelines

Small backend and data workflows that keep projects usable end to end.

Experiments

Computer vision, speech processing, and mobile ideas tested as working prototypes.

Selected Work

Projects written with the tradeoffs left in.

Each case study explains the problem, what I tried, what I measured, and where the constraints shaped the implementation.

All projects

Products

20k

History

3 yrs

Zero demand

85%

FeaturedForecastingLightGBMTweedie

Intermittent Demand Forecasting System

A two-stage machine learning system for intermittent demand forecasting on daily product-level sales data.

Case study

Runtime

40m -> 4m

Rows

3M

Data

~1 GB

OptimizationNumPyPandas

ML Pipeline Runtime Optimization

Optimized a large-scale ML feature engineering and inference pipeline by reducing unnecessary computation, memory bloat, and Python-level bottlenecks.

Case study

Branches

2

Selector

Confidence

Output

Pose

Computer VisionRAFTORB

Hybrid Visual Odometry System

A confidence-aware visual odometry system combining deep learning-based optical flow and classic computer vision.

Case study

Tools

Tools I reach for when the problem calls for them.

PythonLightGBMNumPyPandasScikit-learnPyTorchHugging FaceWhisperTypeScriptNext.jsFlutterFirebasePrisma

Notes

Short notes from the build log.

Read notes
WhisperNLPDeployment

Building Self-Hosted Whisper Pipelines

Whisper becomes more useful when transcription is treated as one stage in a larger NLP system.

Computer VisionVisual Odometry

Hybrid Visual Odometry: RAFT + ORB

A confidence selector can turn deep and classical visual odometry branches into a more robust research architecture.

OptimizationNumPyPandas

Vectorization vs Python Loops in ML Pipelines

The fastest pipeline improvement is often moving repeated Python-level work into vectorized array operations.