Projects

• Aspect-Based Sentiment Drift Detection — Detecting sentiment drift in Amazon reviews using PyABSA and LSTM-Autoencoder.
• PowerGraph: Cascading Failure Prediction — Graph Neural Network pipeline predicting cascading blackouts in power grids.
• Diabetes Risk from Survey Data — Diabetes screening through CatBoost and clustering on the BRFSS survey dataset.
• KV Cache Compression Benchmark — Benchmarking 7 KV cache eviction methods on NIAH retrieval.
• Few-Shot Tool Wear Prediction — Physics-informed Multi-Target GPR for milling cutter wear prediction from 6 training trajectories.
• Student Success Predictor — Multiclass classification pipeline predicting student outcomes from OULAD with XGBoost and SHAP explainability.
• Foundation Model Survival Analysis — Evaluating frozen time-series foundation model embeddings as feature extractors for survival-analytic prognostics across heterogeneous degradation datasets.
• Quotation Processor — Quotation automation using n8n, FastAPI, and semantic SKU matching.
• Bilkent Honor Search — Track Bilkent University student honor list achievements across semesters.
• Smart Screen — Resume screening tool reducing HR candidate review from 30 minutes to under 1 minute per batch using RAG Fusion over 100+ resumes.
• SVQ.AI — LLM-powered RAG system using LightRAG hybrid retrieval for multi-format document ingestion and sub-second regulatory queries.
• Campus Connect — Full-stack marketplace enabling 13,000+ Bilkent students to buy, sell, donate, and borrow items with real-time auctions.
• MIPS Assembly Pipelined Processor — 5-stage pipelined MIPS processor in Verilog deployed on Basys3 FPGA.
• PIC32 Microcontroller — Motor control and seven-segment display programming on a PIC32 microcontroller using C.
• Till The Last Drop — Tower defense game with a physics engine for NPC pathfinding, collision detection, and rendering.