ProgrammeMSc Computer Science ETH Zürich, since 2026
SpecializationMajor: Secure and Reliable Systems Minor: Machine Intelligence
ExperienceTelemetry, robotics, full-stack systems
DirectionSecurity for robotics and physical AI
Alişer UlaşGöner.
I build software for systems that work in the real world.
Live telemetry for an underwater glider, robotics software for a sorting robot, web infrastructure for a founder community. Next: security for robots and physical AI.
01Illustration: an underwater glider on its sawtooth path.
02Illustration: a robot arm picking items from a conveyor.
03Illustration: the co-founder matching method.
04Illustration: a vague word refined into confirmed criteria.
05Schematic: an agent calls tools that query each data source.
§ 02
Work
01
Project NAUTILUS
Software Team MemberARIS, Academic Space Initiative SwitzerlandSep 2024–Feb 2026
Web interfaces for live telemetry from an unmanned underwater glider.
Responsible for the telemetry frontend, the interface the team used to monitor live data during field operations.
Developed the frontend in Vue, TypeScript and HTML/CSS for real-time monitoring of sensor data.
Containerized development and deployment with Docker.
Vue
TypeScript
HTML/CSS
Docker
With ARIS
Context
NAUTILUS is the ARIS project to build the organization’s first autonomous underwater glider. The glider moves by changing its buoyancy and gliding on wings instead of using propellers. It is designed to collect data for polar and climate research at lower cost and with less environmental impact than ship-based surveys.
I was responsible for the telemetry frontend: how the glider’s live data was visualized, and the interface design the team relied on during dive tests.
How it works
IllustrationHow an underwater glider moves
A glider changes its buoyancy to sink and climb, turning vertical motion into forward travel along a sawtooth path.
A web interface in Vue and TypeScript that receives live telemetry and shows sensor data in real time. During dive tests, the whole team followed the glider through it, including the test in which the glider completed three successful dives to a maximum depth of 22 m.
How I worked
I designed the interface as well as building it, laying out the live data so the team could follow a dive at a glance. I also containerized development and deployment with Docker so everyone on the team worked with the same setup.
Takeaways
Software for field tests is read under pressure. The interface had to show the essentials at a glance to a team of engineers from different disciplines, while the glider was in the water.
02
Project MIDAS
Robotics EngineerRobotic Sorting, ETH Robotics ClubFeb–Sep 2026
Robotics software, data and vision research for an autonomous robot that sorts PET and aluminium waste.
Converted the robotics software stack from ROS1 to ROS2.
Collected and edited training data, and maintained the project database.
Researched robotic vision approaches and reached out to companies about possible industrial projects.
ROS2
Python
YOLO
Docker
Data
Context
Project MIDAS (Material Identification and Autonomous Sorting) is the ETH Robotics Club’s robotic sorting project, which grew out of the Autonomous River Cleanup project. The team turned a waste-analysis station into a compact conveyor testbed with a robotic arm. The system detects PET and aluminium objects, grasps them, checks each grasp and drops the object into the right bin, without manual target selection.
My role
My work covered five areas: collecting and editing training data, converting the software from ROS1 to ROS2, maintaining the project database, researching robotic vision, and reaching out to companies about possible industrial projects.
How it works
SchematicThe sorting cycle
Classical control moves the arm; a learned policy handles the final grasp.
My work sat underneath this cycle: the ROS2 software it runs on, the data used to train it, and research into the vision side.
How I worked
The main technical problem I took on was converting the software from ROS1 to ROS2. I also looked outward: alongside the engineering, I reached out to companies about possible industrial projects for the system.
Takeaways
Much of the hard work in robotics sits between the components. Migrations, data and infrastructure decide how quickly a team can try anything new.
03
ETH Entrepreneur Club
Full Stack Engineer, IT CommitteeFeb–Sep 2026
Frontend and database work across the club’s web infrastructure.
Responsible for the frontend: maintained and updated the design across event sites and internal tools.
Handled database changes for the club’s web applications.
Built a Python co-founder matching algorithm for a 45+ attendee event, with fewer than 7% repeat encounters across two rounds.
TypeScript
Python
Databases
Period
Feb–Sep 2026
Stack
TypeScript, Python
Scope
Frontend and database across club sites and tools
Context
The ETH Entrepreneur Club is a student-run initiative at ETH Zürich that brings entrepreneurial students together through startup events and a founder community.
My role
Within the IT Committee, I was responsible for the frontend and for database changes across the club’s web infrastructure. I maintained and updated the frontend design throughout.
Side project
For a co-founder event with more than 45 attendees, I wrote a Python algorithm that scores every pair of attendees, weights rare industry overlaps higher, and improves the groups through swap optimization with several restarts. It formed balanced groups across two rounds with fewer than 7% repeat encounters.
IllustrationHow swap optimization forms groups
TechnicalBusinessDesignMet in round 1
Each step swaps two people and keeps the swap only if the groups improve. In round two, pairs who already met count against a group.
04
Bachelor’s Thesis
Interactive Visualization & Intelligence Augmentation Lab, ETH ZürichFeb–May 2026
A web workspace that helps people turn a vague word like “interesting” into concrete criteria for ranking texts, with an LLM proposing and refining the options.
Built the full system: an HTML frontend and a Python backend.
Used an LLM to generate candidate criteria and grouped them by meaning.
Designed rules that decide when the system offers new suggestions.
Python
HTML
LLMs
Clustering
Title
Interactive Clarification for Evolving Attribute Spaces in Multi-Criteria Text Ranking
Lab
IVIA Lab, ETH Zürich
Grade
5.5 / 6
Context
People often judge texts with broad words like “interesting” or “clear”. These words mean something between people, but a system that has to rank texts cannot act on them. The thesis asks how an LLM and a visual, evolving set of candidate attributes can help a user turn such a word into concrete, confirmed criteria.
My role
This was my individual bachelor’s thesis at the IVIA Lab. I designed the workflow, built the full system and wrote the report.
How it works
ScreenshotThe workspace
The workspace at the start of a session: chat on the left, candidate attributes grouped into named clusters on the right.
The user types a descriptor. The system asks the LLM for about 24 candidate attributes, embeds and clusters them, and names each cluster. The user confirms, rejects, moves or merges candidates, either directly or through chat, and every operation is recorded. When the user’s behavior matches one of five rules, the system offers to generate new candidates in a direction the user chooses.
ScreenshotEnd of a session
Confirmed, rejected and user-added attributes, a plain-language reflection, and a preference profile for a later ranking step.
How I worked
I prototyped four interaction designs before settling on clustered candidates that the user corrects, with a chat and rule engine on top.
Takeaways
People are better at reacting to concrete options than at listing what they want up front. Designing around that shaped the interface more than any model choice.
05
conteLLigence
Colosseum HackathonETH Entrepreneur ClubMay 2026
An AI agent that diagnoses e-commerce marketing performance.
Built the frontend that surfaces the agent’s diagnoses and supporting metrics.
Implemented a retrieval layer that grounds the agent’s answers in the project’s data sources.
Role
Frontend and retrieval layer
Stack
AWS Bedrock, Lambda
Sources
Shopify, Instagram, Google Analytics
How it works
SchematicSystem outline
The agent calls Lambda tools that query each data source. Filled boxes mark my part.
§ 03
About
I’m an MSc Computer Science student at ETH Zürich, majoring in Secure and Reliable Systems with a minor in Machine Intelligence. Most of my work so far has been software that meets hardware: telemetry for an underwater glider, data and infrastructure for a sorting robot. Next, I want to work on making robots and physical AI systems secure. Working with student founders made me curious about the other side of the table. I’m now learning how technology companies are evaluated and funded, with the goal of becoming an investor myself one day.
Outside of work, I swim, train kickboxing, play the piano and the saz, and write poetry, some of which has been published.
Education
MSc Computer ScienceETH Zürich · since 2026 · Major: Secure and Reliable Systems · Minor: Machine Intelligence
BSc Computer ScienceETH Zürich · 2022–2026 · Thesis at the IVIA Lab
