About Me
I am a highly motivated Electrical and Electronics Engineering student at Bilkent University, specializing in AI-driven systems, computer vision, and robotics. My expertise spans reinforcement learning for energy and financial markets, deep learning for image processing, and FPGA-accelerated computer vision. Through my research at Bilkent's LiRA and GDN Labs, I have developed advanced robotic simulation environments, task-and-motion planning algorithms, and game-theoretic multi-agent systems.
Beyond academia, I am actively involved in interdisciplinary projects, including autonomous systems, financial algorithmic trading, and human-robot interaction. I have a passion for problem-solving in complex, non-linear systems and aim to contribute to cutting-edge research in robotics and AI. Additionally, my experience in theatrical performance and vocal training has refined my communication and leadership skills, allowing me to adapt to dynamic team environments.
I aspire to further my expertise in AI, robotics, and space technology through research and industry experience, ultimately working towards innovations that bridge intelligent automation with real-world applications.
Education
Bachelor of Science in Electrical & Electronic Engineering
Bilkent University
3rd Year Student
3.30/4.00 GPA
Skills
Programming
- Python
- C/C++
- MATLAB
- R
- Spice
- Git
- Linux
- ROS2
Electronics
- Circuit Design & PCB Layout & Simulation
- Microcontrollers
- FPGA Programming
- Soldering & Prototyping
AI & Robotics
- Machine Learning & Deep Learning
- Computer Vision
- ROS2 (Robot Operating System)
- Sensor Integration
- Control Systems
- Reinforcement Learning
Finance
- Financial Modeling
- Data Analysis
- Algorithmic Trading
- Risk Assessment
- Quantitative Analysis
Projects
Energy Market Optimization
Developed an AI-driven energy distribution optimization system using reinforcement learning to balance supply and demand in power grids. Implemented multiple RL agents (PPO, SAC) to optimize energy allocation while considering renewable sources, price volatility, and emissions. The system includes a comprehensive market simulator with visualization tools for demand-supply analysis.
View Project
Diffusion Model for Scene Understanding
Developed a comprehensive system that automatically interprets 2D images to build scene graphs describing objects, relationships, and contextual cues. The system combines state-of-the-art object detection (YOLOv8/DETR), image captioning (BLIP), and natural language processing techniques to extract relationships between objects and construct detailed scene graphs.
View Project
Autonomous VSLAM Robot
Developed a Visual-Inertial Simultaneous Localization and Mapping (VISLAM) robot capable of autonomous navigation in bounded environments. The system integrates ROS2, a Raspberry Pi 4, and custom FPGA acceleration for image preprocessing to achieve real-time performance. The FPGA hardware accelerator, written in VHDL, filters images into grayscale and achieves a 22% improvement in computation speed.
Power Grid Estimator / Solar Forecaster
Developed a sophisticated machine learning system using LSTM neural networks to forecast solar activity and space weather conditions critical for power grid stability and satellite operations. The system analyzes historical solar data (GOES X-ray flux) and solar wind parameters to predict future conditions with multi-horizon forecasting capabilities up to 12 time steps ahead.
View Project
Satellite Image Super-Resolution
Developed a deep learning-based solution for enhancing the resolution of satellite imagery using state-of-the-art super-resolution techniques. Implemented multiple models including ESRGAN (primary), SRCNN, and FSRCNN with 4x resolution enhancement capabilities. The system includes both a command-line interface for batch processing and a web application for interactive use.
View Project
Multi-Agent Resource Competition Simulation
Designed a Dec-POMDP environment under partial observability to simulate thief-villager interactions under restricted resources and used PPO for decentralized policy training followed by centralized application.
View Project
EchoArm
Designed a large robotic arm engineered to replicate the movements of a smaller, manually operated arm, with up to 4 DoF. Uses KL25Z development board as the main processing unit.
View Project
Specialized Chat-Bot
Trained and implemented a basic specialized Chatbot that leverages retrieval augmented generation (RAG) using GPT-4 LLM, OpenAI embeddings, ChromaDB, and LangChain.
View Project
Plate Recogniser
A plate recognition system that uses Haar Classifiers as a base to find plates in a prerecorded video feed. Has adjustable parameters that specifies the maximum amount of classifications allowed in a frame.
View Project
HFT/Normal Algorithmic Trading Pipeline
Designed a trading app which interfaces with the algorithmic trading pipeline, which trades for you with your desired risk parameters, custom strategies, portfolios, and currencies in real time.
View Project
Market Forecaster
Designed a pipeline integrating reinforcement learning with time-series analysis to extract trends and correlations across stocks and cryptocurrencies. Also working on news sentiment analysis for black swan detection.
View Project
ALU
Designed and implemented a modular ALU that can do basic arithmetic and shifting operations. Is adjustable for extra operations. Was implemented on a BASYS3 FPGA Board.
View Project
FPGA-to-Computer Ethernet Network
Designed a 10/100T Ethernet capable FPGA network compatible with computer connections using Nexys A7-100T FPGA development board. Uses IPV4 protocol and responds to specific UDP frames.
View Project
Quadruple Servo Tester & Fault Detector
Designed and implemented a quadruple servo driver with 4 potentiometers for rotation control using NXP KL25Z4 MCU board. Includes an internal OPAMP buffer circuit for detecting faulty servos.
View Project
STM32F103C8T6 Custom Communication PCB
Designed an STM32F3 MCU board with RS232, 4-channel DAC, 2-channel ADC, and USB support with an ST-Link programming interface. Communicates with MCP4728 DAC IC via I2C.
View Project
Thermocouple Instrumentation Amplifier Controlled Heater
Designed, simulated, tested, and implemented a custom analog thermocouple instrumentation amplifier-controlled heater circuit on a one-sided PCB as coursework for Bilkent University.
View Project
Automatic Trading Bot
An automated trading bot which can be both deployed in a test server or official trading servers via Cloud. Uses a default MA&RSI strategy as base. Strategy used is adjustable. Currently inactive.
View Project
Regression Model
Designed and implemented a curve fit model which uses polynomial regression, linear regression, exponential fit, logarithmic fit, and spline interpolation. Can be used by supplying html, css, or excel files.
View Project
Customizable Maze-Based Manipulation Puzzle Simulation
Developed a simulation environment for testing robotic tasks in maze scenarios. Features include a custom maze generator, inverse kinematics solving, and JSON-based scenario sharing.
View Project
Optimized Bottleneck RRT
Enhanced Rapidly-exploring Random Tree (RRT) algorithm with subgoal sampling, bottleneck detection, path smoothing, and optimized rewiring to improve pathfinding in complex environments with obstacles.
View Project
Swetlana Maze Solver
Designed a tool that solves pick&place problems with K-order Markov Optimization and bottleneck RRT using RAI API for Python.
View Project
MetaTrader5 Abstraction Layer
Wrote an abstraction layer for the Meta Trader 5 Python integration library. Currently, only client-side requests are available. Flask integration for broker-side requests is in progress.
View Project
Data Analyzer
A program that can explore any finite data set and fit an appropriate nth degree polynomial. Visualizes the fitted polynomial against the data points. Uses Polynomial Regression to fit a function to the dataset.
View Project
FPGA UART Loopback Server
Implemented a UART server between a Nexys A7-100T FPGA development board and a computer. Uses a Microblaze-based system with AXI4 streams and predefined UART IPs.
View Project
STM32F103C8T6 RS232 Driver
Wrote a user-friendly RS232 driver using USART peripherals for STM32F3 chips.
View Project
STM32F103C8T6 DAC Driver (MCP4728)
Wrote a user-friendly STM32F3 driver for MCP4728 4-channel DAC. I2C interface is used to communicate with the DAC. Functions included mimic the functions in the original datasheet.
View Project
Custom LDO Converter Circuit
Designed, simulated, tested, and implemented an LDO converter on a breadboard as coursework for Bilkent University.
View Project
Diode Differential Temperature Sensor
Designed, simulated, tested, and implemented a Diode-characterized differential temperature sensor as coursework for Bilkent University.
View Project
LDO Converter PCB
Designed a DC/DC isolated LDO converter IC circuit with Altium. It transforms 5V to 3.3V.
View Project
TRC-11
Implemented, tested, and simulated a TRC-11 transceiver radio as lab work at Bilkent University.
View Project
Retro Video Game
A 2D Mario-like game I developed as a High Schooler. Was written using Unity Engine and C#. Used the internal tools of Unity to design levels and create pixel art.
View Project
SPH Fluid Simulation
Developed a 2D Smoothed-Particle Hydrodynamics (SPH) fluid simulation in C++ with OpenGL rendering. Features include real-time parameter tuning via ImGui for adjusting gravity, viscosity, pressure, particle count, smoothing radius, and damping coefficient. Implemented multiple SPH kernels for accurate fluid physics.
View Project
Playlist Recovery Agent
Developed a machine learning pipeline for analyzing and reconstructing music playlists from shared accounts. The system preprocesses music data, performs clustering, and uses RandomForestClassifier models optimized with Optuna to predict user preferences and reconstruct personalized playlists.
View Project
Federated Learning for Decentralized Robots
Developed a framework enabling multiple robots to collaboratively train machine learning models without sharing raw data. The system preserves privacy by exchanging only model parameters or gradients, addressing privacy and bandwidth constraints in multi-robot learning scenarios. Implemented and tested on image classification tasks with simulated robots.
View ProjectExperience
Engineering Intern
Tech Company
May 2022 - August 2022
- Contributed to the development of embedded systems for IoT applications
- Performed testing and debugging of electronic circuits
- Participated in team meetings and presented project updates to stakeholders
Research
Task and Motion Planning with Gaussian Search
Bilkent University - Learning and Intelligent Robotics Laboratory (LiRA)
Currently investigating Gaussian and Bayesian search algorithms for Task and Motion Planning (TAMP) in robotics. The research focuses on developing efficient search strategies for complex manipulation tasks in uncertain environments.
Key Areas:
- Gaussian process optimization for motion planning
- Bayesian search algorithms for task planning
- Integration of probabilistic methods in TAMP frameworks
- Optimization of search strategies for robotic manipulation
Traffic Control Optimization
Bilkent University - Game Design and Development Laboratory (GDN)
Researching optimal controllers and policies for traffic congestion mitigation, with a focus on single-lane ring road setups. Developing heuristic-based simplifications of current trending techniques to improve computational efficiency while maintaining effectiveness.
Key Areas:
- Heuristic optimization for traffic flow control
- Single-lane ring road traffic management
- Simplified control policies for congestion mitigation
- Performance analysis of traffic control strategies
FPGA Acceleration for Computer Vision
Independent Research
Investigating the use of FPGA hardware acceleration for real-time computer vision applications in robotics. Focusing on optimizing image preprocessing for SLAM algorithms.
Key Contributions:
- Implemented grayscaling and filtering algorithms on Artix-7 FPGA
- Achieved 22% improvement in computation speed for image preprocessing
- Developed modular design compatible with various filter types
Reinforcement Learning for Financial Markets
Independent Research
Investigating applications of reinforcement learning algorithms for financial market prediction and algorithmic trading. My research focuses on combining traditional time-series analysis with modern RL approaches to create robust prediction models.
Key Areas:
- Time-series forecasting using LSTM and transformer architectures
- Market regime detection and adaptive trading strategies
- Risk-aware reinforcement learning for portfolio optimization
- Combining sentiment analysis with price action for improved prediction accuracy