Hello, I'm
Pursuing research at the intersection of mathematics, machine learning, and physical intelligence, aiming to build the next generation of intelligent systems that can interact with the real world.
And I prefer principled models over voodoo constants.
Computer Science & Mathematics undergraduate at the University of Manchester. Right now that means working toward becoming an AI researcher, research engineer, and embodied AI researcher — getting intelligent systems to actually hold up against real-world problems, not just clean benchmarks. The long-term goal hasn't moved: building the next generation of intelligent systems.
I co-authored a research paper on nonlinear time-series forecasting for air pollution using chaos theory and statistical modelling, presented at the CHAOS 2025 International Conference in Greece. The work deepened my expertise in dynamical systems, probability theory, and data-driven analysis.
At Manchester Stinger Motorsports, I own simulation development and validation across the Formula Student autonomy team's perception, planning, and control interfaces — diagnosing controller failures from test logs and refactoring a bicycle-model simulator into a ROS 2 closed-loop environment.
During my internship at BASARSOFT, I engineered large-scale data validation pipelines for 21 million geospatial records, improving data integrity across 3 systems by 40%. I also completed an intensive ML, AI & Data Science summer programme at Sabancı University.
Further out, my interests widen into neuromorphic and biological computing and hardware–software co-design, with quantum computing somewhere past that horizon. But the near-term plan stays concrete: AI research, research engineering, and embodied AI — systems that work in the real world before anything more speculative.
Modelling system sensitivity and chaotic dynamics across environmental datasets spanning 5 cities and 3 years of observations. Built reproducible Python pipelines covering automated preprocessing, statistical feature extraction, and visualisation across 5 dataset variants — increasing iteration efficiency by 40%.
Chaos and learning sit closer than most people think. Both ask how small perturbations propagate through nonlinear systems; both depend on the geometry of high-dimensional trajectories. The paper was an excuse to live inside that question for fifteen months.
"Real systems don't apologise for being nonlinear. The challenge is to model them honestly."
Pulling these intuitions into representation learning and embodied control — where the dynamics aren't a dataset, they're the agent.
Sole developer. Production AI legal assistant — native iOS app (SwiftUI) and FastAPI backend querying 15+ live Turkish government/court APIs via MCP, synthesising cited answers with Claude. Full auth, security-audited, admin panel, Sentry monitoring.
Built a simulation-first PX4/ROS 2 quadrotor stack and a 9,216-episode Isaac Lab study. Bounded residual PPO achieved 1,023/1,024 severe paired successes (99.90%) with zero collisions, then completed one guarded Gazebo/PX4 navigation-and-landing mission with classical fallback.
Built image-based item detection and estimated resale pricing for a secondhand assistant. Integrated a Unitree Go2 robot to navigate to saved viewpoints, capture objects, and feed the same review workflow used by phone and Meta glasses.
Built the neural networks for a hackathon benchmark predicting peptide–HLA dissociation half-life. A positional-BLOSUM MLP regressor over 27,031 pairs beat frozen ESM-2 protein-language-model features — then I re-ran ESM-2 through the same architecture to prove the gap was the network head, not the representation.
Context-aware retrieval & ranking pipeline for music. Tag-based candidate generation, lightweight ranking on weather / season signals, and a low-latency PostgreSQL schema designed for real-time scoring.
Two-node Raspberry Pi Pico telemetry prototype for search-and-rescue rover experiments. A Scout samples ultrasonic range and BMI160 gyroscope data over nRF24L01; an Anchor feeds a PyQt6 dashboard with live instruments and CSV logging.
BSc (Hons) Computer Science & Mathematics — expected First Class.
Modules: Linear Algebra · Probability · Statistics · Data Science. Research focus on AI, research engineering, and embodied AI.
Intensive modules in Machine Learning, AI & Data Science.
International Baccalaureate Diploma · HL: Mathematics Analysis & Approaches, Physics.