Ahmet Efe
Building systems.
Thinking in images.
I'm a software engineer and an M.Sc. student at ETH Zürich, focused on computer graphics and visual computing. I like understanding how things work—from the systems behind a product to the geometry and light behind an image.
At Midas, I build backend systems for global markets. Away from the keyboard, I explore Zürich, Istanbul, and the places between them with a camera.
Grade this page
Make it feel like you
Pick an accent
Your palette, across the page. Photos stay original.
Journey
A shared timeline of study and practice. Follow the overlaps, then explore the chapters behind them.
EducationThe foundations
Work & practicePutting it into practice
Midas · Global Trade
Jun 2026 — now
I build backend systems for Global Trade at Midas, supporting US and EU markets with Java, Spring, and GraphQL. The stack spans event-driven messaging with Kafka, Redis, and Kubernetes deployments through ArgoCD. I joined as an intern and moved into a Software Engineer role in September, continuing part-time alongside my M.Sc.
ETH Zürich
2025 — now
My focus is Visual & Interactive Computing: how images are formed, how visual information is interpreted, and how shapes and motion can be represented computationally. Coursework spans computer graphics and vision, shape modeling and geometry processing, computational models of motion, and computational intelligence. Big Data adds a complementary perspective on processing information at scale.
Private tutoring · Zürich
Jan 2025 — Jun 2026
I tutored two Kantonsschule students, primarily in mathematics, with additional support in physics, chemistry, and biology. Sessions centered on breaking down difficult concepts and working through problems step by step, adapting the explanation to each student and connecting abstract ideas to concrete examples.
Outlier AI · Remote
Dec 2024 — Jun 2026
I developed and evaluated prompts for AI chatbots, focusing on mathematical reasoning, accuracy, and model safety. The work involved assessing how models approached mathematical problems and collaborating with cross-functional teams to improve the quality of their reasoning and responses.
HABEE Solutions · Zug
May — Jul 2025
I maintained and improved internal software at HABEE Solutions and built tools to automate repetitive workflows. The internship combined work in existing codebases with practical automation, turning manual processing steps into software-supported workflows and reducing the time spent on routine tasks.
ETH Zürich
2021 — 2025
A broad foundation in computer science, from algorithms, probability, and numerical methods to systems programming, computer architecture, networks, and databases. I explored compiler design and rigorous software engineering alongside machine learning, visual computing, human–computer interaction, and information retrieval. Work in web engineering and FPGA design connected these ideas to software and hardware; my thesis brought the visual side together in a computational embroidery pipeline.
Bachelor thesis
Computational Design Lab, ETH Zürich · 2024–2025. Image segmentation for directionality-aware embroidery.
Tiny color islands and fragmented regions interrupt the direction and flow of embroidery.
Combine neighboring superpixels using color variation and boundary evidence, then organize the regions into a hierarchy.
Export layered polygons and direction hints for the next stage of the embroidery pipeline.
A computer-graphics pipeline that prepares images to become directional thread.
A region that looks plausible on screen is not necessarily a good region to stitch. Fragmented color islands create unwanted jumps; too many tiny regions break up smooth flow. My thesis explores how segmentation can preserve boundaries, internal color variation, and nested detail for two-tone, directional embroidery.
Starting with SLIC superpixels, I greedily merge neighboring regions using perceptual color, PCA-based color variation, and boundary evidence. The resulting polygons form a containment hierarchy, exported as one LabelMe JSON per depth with automatic direction hints for the downstream embroidery system of Liu et al.
Explore the thesis on GitHubFour cues contribute to a shared affinity. A single weak factor can make a merge too costly. Adjust the factors to explore the decision.
Mean colors in CIELAB space: larger perceptual differences reduce affinity.
Alignment of the dominant color-variation axes, with a penalty conditioned on their lengths.
A factor based on the ratio between the smaller and larger color-variation lengths.
Canny edge evidence along the shared boundary: strong edges discourage merging.
Illustrative factor values, not a live image segmentation. The product and log-cost follow the thesis; 2.80 is one example threshold. After each accepted merge, region features and neighboring graph costs are recomputed.
Cleaned polygons are organized by containment, so background regions can be stitched before their nested details. Shape-derived inner chords provide initial direction hints, connecting image segmentation to the next stage of pattern generation.
Qualitative comparisons with K-means, raw SLIC, and simulated detector masks showed more cohesive regions and smoother stitch-flow previews. Fine textures, image-dependent thresholds, and robust polygon conversion remain challenges; automatic direction hints can still benefit from manual refinement.
Photography
Small observations from mountain paths and city streets. A collection of light, architecture, and everyday details, photographed with my Sony A7C and iPhone.
12 of 48 photographs
Camera gear
Shown at true relative size Hover or tap to explore
Contact
Have a problem worth looking at twice?
Let's talk.