Hi, I’m Berkay Arda.

This team project focused on developing an intelligent surveillance system that leverages real-time weapon detection and human behavior analysis to improve public safety. My role involved training and evaluating object detection models (YOLOv8–v12) using custom-labeled CCTV datasets and implementing a VideoMAE-based architecture for suspicious behavior recognition. I integrated these models into a modular pipeline and built a mobile-friendly notification system with Firebase to deliver alerts instantly.

This interdisciplinary research project explores how artificial intelligence and bioengineering can be combined to create fire-resistant forest ecosystems. As part of a multidisciplinary team, I contributed to an early-warning system that analyzes environmental and vegetation data to detect wildfire risks using Random Forest models.

We also proposed integrating AI-based risk analysis with genetically engineered tree species that exhibit fire-resilient properties, plus a monitoring infrastructure using thermal cameras and drone surveillance.

Built a Retrieval-Augmented Generation (RAG) system that answers questions strictly from a book’s content. The pipeline parses and cleans documents, chunks text, creates embeddings (Sentence-Transformers), stores vectors in FAISS, and retrieves top context passages for the LLM. Includes retrieval gating to reduce hallucinations and improve answer faithfulness.

• 📄 PDF/Text parsing, cleaning, and chunking (overlap-based)
• 🧠 Embeddings with Sentence-Transformers + FAISS vector search
• 🧭 Retrieval gating for WHY/HOW style queries and relevance control
• ✅ Prompting strategy: “Answer only using retrieved context”

Developed a comprehensive event scheduling and notification platform using C# (WinForms) and ASP.NET MVC, integrated with an SQL database. Implemented key features such as user registration, role-based access control (admin/staff/user), real-time event editing, and automated notification alerts.

• 👤 Role-based user registration (Admin, Staff, Participant)
• 📅 Event creation, editing, and scheduling
• 🔔 Notifications for updated/canceled events
• 📈 Filters and sorting for event tracking

EasyAssign is a lightweight Windows application developed to help educators evaluate programming assignments efficiently. It supports multiple languages and provides a structured workflow from submission to grading.

• 📁 Automatic ZIP processing from student submissions
• 💻 Compilation and execution
• 📊 Output comparison against expected results
• 🛠️ Error reporting for compile/runtime issues

This project is a 2D action-adventure game developed during my internship at Teracity Software Technologies. The goal was to design a playable core game loop while applying clean code practices and object-oriented design principles.

• Character Movement & Combat: Implemented smooth 2D movement and combat mechanics, including melee (sword) and ranged (bow) weapon systems.
• Enemy AI: Designed multiple enemy types with proximity-based behaviors and attack logic.
• Wave-Based Spawning: Enemies spawn in increasing waves, with a boss encounter triggered after every 10 kills.
• Inventory System: Fully functional inventory allowing players to collect, use, and drop items.
• User Interface: Developed optimized UI elements including score tracking, wave indicators, inventory display, and health feedback.

The project focuses on maintainable architecture, reusable components, and performance-aware gameplay systems, serving as a solid foundation for future extensions such as new levels or enemies.

🔗 View on GitHub

Kormos: The Beginning is a 3D supernatural detective shooter developed for the SE320 – Introduction to Computer Games course. The game blends psychological storytelling with mechanics-driven gameplay, focusing on time manipulation, teleportation, and strategic combat.

The player controls Kormos, a detective suffering from amnesia who dives into his own subconscious to recover lost memories and confront manifestations of his past.

• Enemy AI & Boss Design: Implemented multiple enemy types including Baby Rats and Priests, as well as all boss encounters (Rat King, Zangoc, Cutie Devil).
• Core Mechanics: Time slow, teleportation & stun, clone (time remnant) system, dash in 8 directions, and ranged combat.
• Level Design Contribution: Designed and implemented the final level with boss encounters.
• Iterative Game Design: Mechanics refined based on feedback — procedural generation and grappling hook removed in favor of teleportation and time-based abilities.

The final build includes a tutorial level, two playable levels, multiple enemy types, three bosses, and a minimalist UI focused on clarity and immersion.

🔗 View on GitHub

A playable Tower Defense prototype developed for a case study. The game runs in a 3D environment while using 2D sprite characters through a billboard technique. Enemies follow a defined path using NavMesh and attempt to destroy the base. The project focuses on a clean and extendable gameplay loop with wave progression, boss waves, scoring, and skill-based mechanics.

• Core loop: enemies spawn in waves, follow a path, and attack the base at the end.
• Wave system: UI-based wave tracking, 6s inter-wave wait, and early start with F.
• Boss waves: bosses spawn on 5th / 10th / 15th waves, then normal flow continues.
• Scoring & skills: +10 score per kill; at 200 score you can use Heal (E) and Arrow Rain (Q) with cooldown + UI.
• Player: WASD movement, dash with Shift, and auto-attack the nearest enemy (melee/ranged-ready).
• Performance tests: stress-tested with high enemy counts (200–2000 animated enemies) while keeping smooth FPS.

🔗 View on GitHub

Designed and evaluated a lightweight blockchain-based integrity framework for Wireless Sensor Networks using Contiki-NG and Cooja. The system batches sensor transactions with Merkle trees and performs adaptive block creation based on real-time conditions such as energy level, traffic load, and security alerts—reducing overhead while maintaining strong data integrity.

• 🔋 Energest-based energy monitoring (CPU/LPM/TX/RX) and log-driven evaluation
• 🌲 Merkle Tree batching to minimize on-chain storage and computation
• ⚙️ Adaptive automation layer for block commits (energy/traffic/security-aware)
• 🛡️ Security handling: replay and tamper attack detection & response
• 📡 WSN communication via RPL/UDP with simulation-driven metrics