KEMAL KILIÇASLAN

KEMAL KILIÇASLAN

Super Intelligence Developer

I build projects end to end, from data collection and model training to deployment. I graduated in Mathematics from Kastamonu University in 2022 and have since built 20 production-level projects in ADAS, traffic analytics, medical imaging, fraud detection and on-premise RAG. I taught a short, seasonal AI training at BTK Akademi and created tutorial content for the Ultralytics Platform launch through a brand partnership. I'm currently working on real-world data science and AI solutions.



Education

Kastamonu Üniversitesi

KASTAMONU ÜNİVERSİTESİ FEN EDEBİYAT FAKÜLTESİ MATEMATİK BÖLÜMÜ

2022 Graduate - Bachelor's Degree

Personal Initiatives

Kastamonu Üniversitesi Matematik ve Bilim Topluluğu

KASTAMONU ÜNİVERSİTESİ MATEMATİK & BİLİM TOPLULUĞU


(Founding President)

[December 2021 - June 2022]

 Kemal Kılıçaslan - My speech on the meaning and importance of the day

'14 March π Day'



Türkiye Matematik Kulübü

Türkiye Matematik Kulübü


(Member)

[November 2020 - ]

 I took an active role in the organization team of 6 events, 4 of which were academic and 2 of which were popular, where we addressed more than 1500 people in total.

Talks & Workshops

Niğde Ömer Halisdemir Üniversitesi Bilgisayar Mühendisliği Kulübü

Niğde Ömer Halisdemir Üniversitesi Bilgisayar Mühendisliği Kulübü
(Speaker)

May 15, 2025

At the TeknoKonferans NÖHÜ'25 event organized as part of the 'Education, Collaboration, and Competition for the Software Sector' at Niğde Ömer Halisdemir University, I gave a talk titled 'An Interdisciplinary Career Journey from Mathematics to Artificial Intelligence'.

Experiences

BİLGİ TEKNOLOJİLERİ VE İLETİŞİM KURUMU

BİLGİ TEKNOLOJİLERİ VE İLETİŞİM KURUMU


Artificial Intelligence Instructor

[July 2024 - July 2024]

I gave more than 20 participants face-to-face artificial intelligence training for a week, starting with Python and concluding with machine learning, deep learning and computer vision. Within the scope of the applied curriculum, TensorFlow, OpenCV and GPU accelerated model training demos were used and participants were directed to develop all stages from data preparation to implementation in an end-to-end manner. The training was completed at a 100% rate with each participant attending all five days.



Calorin

Calorin


Computer Vision Developer

[August 2024 - September 2024]

Developed a real-time food volume estimation system using YOLOv8 and OpenCV for pixel-level segmentation of fruits and vegetables, applying geometric calculations to convert segmented areas into calorie estimates with an optimized low-latency inference pipeline validated across diverse lighting and scale conditions.



Ultralytics

Ultralytics


AI Content Creator & Brand Partnership

[March 2026 - July 2026]

Produced a sponsored LinkedIn post and a YouTube tutorial for the Ultralytics Platform launch, demonstrating the full computer vision pipeline (annotation → training → export → deployment) using YOLO on two original real-world datasets.

🔗 Try the Ultralytics Platform




Tech Skills



Python C++ MATLAB Wolfram Mathematica R PostgreSQL Docker HTML5 CSS JavaScript Swift LaTeX

scikit-learn TensorFlow PyTorch OpenCV FFmpeg PySpark BootStrap






PROJECTS



Algorithms & Programming

Kaprekar Constant with Python Programming

This project implements, using only Python code, the process of reaching 6174, the constant discovered by D.R. Kaprekar in 1949, from any four-digit number with at least one differing digit, in at most 7 steps, by repeatedly applying simple arithmetic operations. For three-digit numbers, this number is 495.
Data Analysis & Visualization

Data Visualization of Turkey Population with Plotly

This project implements an interactive visualization of changes in Türkiye's population between 1927 and 2023, using the Plotly and Folium libraries. It maps all 81 provinces and shows the gender distribution in interactive charts, so you can follow how the numbers changed over roughly a century.
Machine Learning

NYC311 Service Channel and Requester Behavior Analytics System

Data: NYC Open Data, 311 Service Requests from 2020 to Present.

This project implements an end-to-end analysis of 22.7M NYC311 service requests with PostgreSQL, R, Python and Spark/Hive. It models how requesters choose between phone, web and mobile, predicts phone requests with LightGBM (AUC 0.883), monitors drift with champion/challenger retraining, and forecasts call-center demand 28 days ahead.

Document and Identity Fraud Detection System

Images: SIDTD (CC BY 4.0), with field boxes and scores added. Application data is synthetic.

This project implements document and identity fraud detection for savings-finance applications. A field-level ResNet-18 finds forged ID fields (ROC-AUC 0.983), and rules plus LightGBM fill the analysts' review queue. Monthly monitoring flags a new device-farm fraud the rules miss, and two candidate rules raise its recall from 12.8% to 82.0%.
Deep Learning

Garbage Classification with Convolutional Neural Network (CNN)

This project implements a CNN model trained on the TrashNet dataset with TensorFlow/Keras. The model learns to tell apart six types of waste (cardboard, glass, metal, paper, plastic, and trash) and reaches 62.2% accuracy. In recycling systems, it can read data from a photo in real time.

Facial Expression Recognition System

This project implements an emotion recognition system that identifies one of seven emotions (anger, disgust, fear, happiness, neutrality, sadness, or surprise) using the VGG19 architecture, with transfer learning and data augmentation techniques applied on the FER-2013 dataset. This particular model achieves an accuracy of 66.33% on the dataset.
Image Processing

Road Lane Lines Detection System

This project implements the Canny edge detection, ROI masking, and Hough transform methods for real-time lane detection, using OpenCV and Python. This is the kind of process that ADAS and autonomous driving systems rely on.

Face Detection and Person Recognition System

This project implements a modular toolkit for face detection and person recognition, built with computer vision and deep learning. It finds faces and can tell who a specific person is, whether you feed it photos, video files, or a live webcam.

Pose Detection with YOLOv8 using Wolfram Mathematica

This project implements human pose estimation by running YOLOv8 inside Wolfram Mathematica. Working from Ultralytics pretrained models, it tracks 17 body keypoints in real time, draws the skeleton over each frame, and comes with tools to analyze how a person moves.

Vehicle Speed Estimation System

This project implements real-time vehicle speed estimation using YOLOv8 and a perspective transform. It follows several cars at once inside a marked region, turns the geometry into a speed in km/h for each one, and saves an annotated video for traffic monitoring.

Traffic Signs Recognition, Vehicle Plate and Person Blurring System

This project implements a real-time recognition system for 24 different regulatory traffic signs commonly encountered in vehicle traffic, using YOLOv11. To protect privacy, it instantly blurs license plates and people. Thanks to this feature, it was developed for autonomous vehicles and GDPR-compliant surveillance systems.

Safety Lane Violation Detection System

This project implements real-time detection and monitoring of the emergency lane using YOLOv12. When a vehicle crosses the safety boundary, an alert is issued and the violation count is updated instantly. Color-coded overlays make each violator easy to identify for inspection purposes.

Vehicle Distance Measurement System

This project implements real-time distance measurement with YOLOv12 for in-vehicle cameras. Using a perspective geometry method, it calculates how far away nearby vehicles are in three lanes (left, front, and right), adjusts its warning thresholds as conditions change, blurs license plates for privacy, and shows color-coded safety warnings.

Road Surface Obstacle and Traffic Warning Object Segmentation System

This project implements real-time, pixel-by-pixel processing of road hazards, obstacles, and traffic warning signs from an in-vehicle camera, using instance segmentation with the YOLO26x-seg model. It enables autonomous vehicles or ADAS systems to see more clearly what is in front of them.

Industrial Safety Gear Detection System

This project implements real-time PPE checks using a custom YOLO26x model trained on the Ultralytics Platform. For each worker it spots, color-coded overlays show whether the helmet, safety vest, gloves, and face mask are on. Accuracy lands at 97.1% mAP50 with 94.6% precision and recall.

Vehicle and Pedestrian Traffic Analysis System

This project implements zone-aware traffic analysis using a custom model trained with YOLO26x on the VisDrone dataset and exported to ONNX format, together with BoT-SORT tracking that I wrote from scratch in C++. The output is displayed live on a dashboard that analyzes people and vehicles in the crosswalk, road, and sidewalk zones.

Vehicle Plates Detection and Reading System

This project implements real-time detection with a custom-trained, transformer-based RF-DETR model and license plate reading with Fast Plate OCR. It detects license plates in traffic footage, reads them, and checks each one against the Turkish plate format, then saves the results to a labeled video for AVPR.

Real-Time Colonoscopy Polyp Segmentation System

Images: HyperKvasir (CC BY 4.0) and CVC-ClinicDB (© Hospital Clinic / Computer Vision Center, Barcelona; research and education only), with model outputs added.

This project implements real-time polyp segmentation for colonoscopy video using a PyTorch U-Net, a frame-quality check, and an IoU tracker that alarms after three consecutive frames. Training on polyp-free frames cuts false alarms on normal colon from 30.6 to 6.2 per minute, and the model scores 0.853 Dice on data from another hospital.

Parking Lot Occupancy and Vehicle Trajectory Prediction System

This project implements parking lot occupancy analysis and vehicle trajectory prediction for drone footage. It finds 400 slots from the painted lines and marks them occupied or empty with a VisDrone-trained YOLO model. Cars in the aisles are tracked with an OC-SORT style tracker, and a dotted line shows where each will be in the next 2 seconds.
Natural Language Processing

On-Premise RAG Document Assistant System

Data: TQuAD and Law No. 6331 on Occupational Health and Safety. Figures from the project's own evaluation.

This project implements an on-premise RAG assistant: an open-weight LLM (Qwen2.5 on Ollama) answers Turkish questions on internal PDFs and cites pages. Semantic search runs on bge-m3 embeddings in pgvector, and role-based access control limits users to allowed documents. On TQuAD, 83.7% of answers are correct when the right chunk is retrieved.





CERTIFICATES