10/2025 — present
Freelance Software Developer
Self-employed · VeraHQ
Production SaaS platforms: multi-tenant architecture, database security, and the deploy pipelines around them.
Profile — Software Engineer · Hatay, Turkey
Software Engineer — Perception, Embedded AI & Full-Stack Systems
I work on measurement, and on what depends on it.
I work where the method only exists on paper — designing the approach, building the hardware and signal chain it needs, and getting it to hold up on real data outside the lab.
Most of what follows happened at the same time as the rest of it. The degree ran from September 2021 to 15 June 2026. The workshops started in 2024 and ran to June 2026. The freelance work started in October 2025 and has not stopped. A bulleted list would hide that, so the next section draws it instead — including the places where the record only gives a year and I will not pretend to a month.
Available immediately
Everything on this page comes from that document. If the two ever disagree, the page is the one that was updated last.
A CV is a list, and a list implies a sequence: degree, then internship, then work. That is not what happened here. The chart below puts all four on one month-resolution axis so the overlap is visible rather than described — and so is the part of the record that is imprecise.
10/2025 — present
Freelance Software Developer
Self-employed · VeraHQ
Production SaaS platforms: multi-tenant architecture, database security, and the deploy pipelines around them.
2024 · 2025
Software Engineering Intern
Beyond Technologies, Ankara
ROS2 perception for unmanned ground vehicles — LiDAR/camera fusion, Cylinder3D on WildScenes, real-time inference on Jetson Orin Nano. Invited back with independent ownership of the perception stack.
2024 — 06/2026
Community Co-Lead & Instructor
Hello World, Adana
20+ hands-on workshops in Java, Python and web development, reaching over 300 students.
The two internship terms at Beyond Technologies are the reason the chart has a comb in it. The record gives the years and not the months, and a solid bar across 2024 would quietly claim twelve months of work that was not twelve months long. The comb says what is actually known: somewhere inside that year.
The teaching sits alongside all of it. Twenty-plus workshops in Java, Python and web development, reaching over three hundred students — that is roughly fifteen people per session (arithmetic, 300 ÷ 20; both figures are written as minimums in the record, so read it as a rough size rather than a rate). Explaining a language to a room that has never seen it is the fastest way I know to find out which parts of your own understanding are decoration.
A five-year programme taught entirely in English, ending with a thesis that became the SHIELD case on this site.
The thesis asks whether a LiDAR can tell that it is being lied to from a single frame, while the vehicle is standing still — the condition under which every motion-based defence I could find stops working. The first version scored 97.1% in simulation and 8.2% on real sensor data. Reporting the first number would have made a better-looking thesis and a detector that fails on contact with hardware, so the collapse became the subject instead. The final system measures 81.6% accuracy on authentic captures.
That episode is the part of the degree I would actually defend in an interview, and it is written up in full on its own page.
A skills list is a claim with no receipt attached. This one carries its receipts: each group is crossed against the four places that actually exercised it, and every filled cell names the specific thing.
| Group | SHIELDBachelor thesis · 2025—26 | Beyond TechnologiesInternship · 2024 · 2025 | PlantSenseSamsung Innovation Campus · 2024—25 | Production platformsFreelance · 10/2025 — |
|---|---|---|---|---|
| Perception & Robotics | Point-cloud handling | ROS2 · LiDAR/camera fusion · Cylinder3D | not used here | not used here |
| Signal Processing | STFT · CFAR · spectrograms | not used here | FFT · acoustic sensing | not used here |
| Machine Learning | Random Forest classifier | Cylinder3D inference on Jetson | not used here | pgvector retrieval |
| Embedded | not used here | Jetson Orin Nano · real-time inference | Pico W · custom analogue circuit | not used here |
| Backend & Data | not used here | not used here | not used here | NestJS · PostgreSQL RLS · PostGIS · Neo4j |
| Infrastructure | not used here | not used here | not used here | Docker Compose · Caddy · GitHub Actions |
Figure 2 — provenance, not proficiency. 11 of the 24 cells are filled — a count, not a score. The empty cells were left empty on purpose; a full grid would not be an honest shape for a five-year record. The teaching role is not a column here because the workshops covered Java, Python and web fundamentals rather than the stack above.
Perception & Robotics
Signal Processing
Machine Learning
Embedded
Backend & Data
Infrastructure
Every project below has its own page, and every page says what was measured and under what conditions. These four figures are the ones I would put in front of a reviewer first.
81.6%
SHIELD accuracy on real sensor captures
38.8ms
per frame — inside a 10 Hz LiDAR budget
69×
event-rate margin, stressed vs. healthy plant
300+
students taught across 20+ workshops
I work on making sensors trustworthy — detecting when a LiDAR is being lied to, and hearing what a plant cannot say.Ali Batuhan Özsürmeli
I am open to engineering roles and to freelance work. If you have a brief, the most useful first message is the constraint you are stuck behind rather than the feature list — that is usually where the interesting part of the problem is, and it is the fastest way for me to tell you honestly whether I am the right person for it.
For hiring: the CV is the same record this page is drawn from, and every measured figure on this site links back to the page that explains how it was measured.
Available immediately