Mert Biricik

Problem: YouTube comment sections are often filled with spam and noise, making it difficult to gauge genuine audience sentiment.

Solution: I built a command-line tool that uses a transformer-based model for sentiment analysis, topic extraction, and spam filtering on YouTube comments. It supports multiple languages and provides clear, actionable insights.

Outcome: The tool effectively cuts through the noise to provide a clear summary of audience feedback, achieving high accuracy in spam detection and sentiment classification.

Problem: Standard search tools cannot answer complex, semantic questions about specific domains like basketball history and player statistics.

Solution: I developed a Retrieval-Augmented Generation (RAG) agent using LangChain and a local LLM (Ollama). The system leverages a specialized knowledge base to understand and answer nuanced, conversational questions about basketball.

Outcome: The agent can successfully answer complex queries that traditional search engines cannot, demonstrating a practical application of domain-specific LLMs.

Problem: Evaluating the performance of fundamental computer systems concepts (like memory allocation and process scheduling) requires standardized and robust testing environments.

Solution: I engineered a comprehensive suite of C/C++ projects to benchmark and demonstrate advanced systems concepts. This included building a custom memory allocator, a process scheduler, a UNIX shell, and a web server.

Outcome: The suite provides a reliable framework for performance analysis and serves as a practical demonstration of core OS and server concepts.