Barış Parlak

Barış
Parlak

AI developer and computer engineering student in Istanbul. I build perception, planning and language systems: a self-driving stack for Turkish traffic, a skin lesion detector, and a Turkish RAG pipeline that never leaves the machine.

Download CV GitHub Email

The road below is a sketch of what Seyir sees. Point at it and you get tracked too.

Projects

Seyir

An autonomous driving stack for Turkish traffic, built from scratch in CARLA 0.9.15.

  1. Perception

    YOLOv11n detects objects, UFLD-v2 finds lanes, Depth-Anything-V2 estimates depth. LiDAR projection fuses them into a 100×100 occupancy grid.

  2. Prediction

    A SocialTransformer forecasts several possible trajectories for every road user.

  3. Planning

    Global A* sets the route, a behaviour state machine picks the manoeuvre, and PPO learns the local decisions.

  4. Control

    An MPC planner in CasADi and IPOPT returns commands in 50 ms or less.

Written in Python and C++ with PyTorch, FastAPI, React and Docker. The road at the top of this page is a small sketch of the same loop.

DeepDerm

Real-time acne detection and remote treatment tracking.

A YOLOv11 model finds lesions in patient photos and grades severity on the Hayashi scale. Doctors review cases in a web panel while patients log their progress in a mobile app.

Trained on data prepared in Roboflow; served by FastAPI and Spring Boot over PostgreSQL, with React and React Native clients.

6 inflammatory lesions Hayashi grade: moderate

Illustration of DeepDerm’s output, not patient data.

TurkRAG

Question answering over Turkish documents with no cloud dependency.

Retrieval combines BM25 keyword search with FAISS vector search, and generation runs locally through Ollama, so sensitive documents stay on your own hardware and within KVKK rules.

Python, LangChain, Sentence-Transformers and FastAPI.

KVKK’ya göre kişisel veriler ne kadar süre saklanabilir?

BM25 finds the words

  1. Saklama ve imha politikası, §4
  2. KVKK madde 7
  3. Aydınlatma metni

FAISS finds the meaning

  1. KVKK madde 7
  2. Veri envanteri rehberi
  3. Saklama ve imha politikası, §4

Fused ranking sent to the local model

  1. KVKK madde 7
  2. Saklama ve imha politikası, §4
  3. Veri envanteri rehberi
Example query. Hover a document to follow it through both rankings.

Experience

  1. –

    AI developer, Caladrius

    Designed a LoRA fine-tuning pipeline for MedGemma 27B that classifies oral epithelial dysplasia and returns structured findings following the WHO 2022 criteria. Set up a two-stage labelling workflow: medical students pre-annotate, specialists revise.

  2. –

    Machine learning developer, Unidev Software Turkey

    Analysed large geospatial datasets and mapped them with Folium. Designed the MyUni MySQL database behind Django REST APIs. Built a personalisation module with collaborative filtering and K-means that raised user engagement by 40%.

  3. –

    Computer Society chair, IEEE İnönü Student Branch

    Formed cross-disciplinary project teams and ran regular algorithm workshops.

Education

  1. –now

    Computer Engineering, İnönü University

  2. –

    Computer Science, Military University of Technology in Warsaw (Erasmus+)

  3. –

    Computer Engineering, Malatya Turgut Özal University

Toolkit

Autonomy and vision
A*, MPC with CasADi/IPOPT, behaviour state machines, PPO, sensor fusion, YOLOv11, lane and depth estimation, PyTorch, LSTM, SocialTransformer
Language models
LoRA fine-tuning, MedGemma 27B, Hugging Face, LangChain, FAISS, BM25, Ollama
Programming
Python, C++ and SQL daily; R and Java when needed
Backend and data
FastAPI, Django, Spring Boot, PostgreSQL, MySQL, Docker, Git, Pandas, NumPy, Folium, React, React Native
Spoken languages
Turkish (native), English (C1), Polish (B1), German (A2), Indonesian (A1)

Terminal

Prefer a keyboard? Type help to see the commands.

Contact

I’m looking for roles in AI, autonomy and data.

barisparlak36@gmail.com

LinkedIn GitHub CV (PDF)